1. sports
2. table tennis and ping-pong

Macroeconomic News and LOB in Foreign Exchange ECN Market
Yusi Tao
Msc in Management Program
Submitted in partial fulfillment
of the requirement for the degree of
Master of Science in Management (Finance)
Goodman School of Business, Brock University
St.Catharines, Ontario
© May, 2015
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Abstract
We investigate the macroeconomic news effect on the dynamics of the limit order
books (LOB) for euro-dollar ECN market in different economic states between Jan.
2006 to Dec. 2009. Using a VAR-STR model on the news surprise, pure news,
aggregated good and bad news, we show that news effects on the LOB dynamics vary
in different states of economy. The LOB dynamics are measured by depth, spread,
slope and volatility. In contract to slope and volatility, depth and spread strongly
respond to news surprise and pure news during recession and expansion. These
characteristics are more affected by aggregated good and bad news during expansion.
News effects are robust to alternative characteristic measures, the different sides of
the LOB and the different levels in the LOB.
Key words: limit order book, depth, spread, slope, macroeconomic news
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Acknowledgements
Foremost, I would like to express my deepest gratitude, special appreciation and
thanks to my supervisor Dr. Walid Ben Omrane. Without his encouragement, patience,
and excellent guidance, I would never have been able to finish my thesis. Also, I feel
extremely lucky to have a supervisor who cared so much about my work, and who
responded to my questions and queries so promptly.
I would also like to express my appreciation to my committee member Dr. Robert
Welch, who provided great advices regarding academic writing. During my
supervisor’s absence, I retained enthusiasm about my work with his timely support
and encouragement.
A very special thanks goes out to Dr. Ernest Biktimirov for the support to make
this thesis possible. Also as my field advisor at the beginning of this program, Dr.
Biktimirov provided me with many useful advice regarding career path.
Lastly, I acknowledge my gratitude to my beloved family and friends. Great
thanks to my best friend Jiahui Wang who always listen to me. I was continually
amazed and inspired by her perseverance and determination. Special thanks to Xinyao
Zhou for helpful advices. I would like to thank for the friendship provided by the
other colleagues of MSc program. And I would like to thank our administration
officers Carrie Kelly, Victoria Steel and Valerie Desimone.
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table of Contents
Abstract .......................................................................................................................... 1
Acknowledgements ........................................................................................................ 1
1.
Introduction ............................................................................................................. 2
2.
Literature Review.................................................................................................... 5
3.
Methodology ......................................................................................................... 10
4.
5.
6.
3.1
Interval Characteristics .................................................................................. 10
3.2
Characteristics of Limit Order Book.............................................................. 13
3.3
VAR with Two-regime Smooth Transition Regression .................................. 19
Data ....................................................................................................................... 27
4.1
Limit Order Book ........................................................................................... 27
4.2
Interval Data................................................................................................... 29
4.3
Macroeconomic News ................................................................................... 31
Empirical Results .................................................................................................. 34
5.1
Characteristics Analysis ................................................................................. 34
5.2
Estimation Results of the Logistic Transition Function in STR model ......... 37
5.3
News Surprise Effects over Business Cycles ................................................. 38
5.4
Pure News Effects over Business Cycles ....................................................... 41
5.5
Asymmetric News Effects over Business Cycles .......................................... 44
Robustness Check ................................................................................................. 45
6.1
Alternative Measures of Characteristics ........................................................ 46
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
7.
YUSI TAO
6.2
Robustness Check of Characteristics ............................................................. 51
6.3
Robustness Check for News Effect on Ask and Bid Sides in LOB ............... 53
6.4
Robustness Check for News Effect on different levels in LOB .................... 54
Conclusion ............................................................................................................ 56
Reference ..................................................................................................................... 58
Appendix A ................................................................................................................ 105
Appendix B ................................................................................................................ 106
Appendix C ................................................................................................................ 107
Appendix D ................................................................................................................ 111
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
List of Tables
Table 1. Descriptive Statistics of LOB ........................................................................ 61
Table 2. Summary Statistics of Characteristics of LOB .............................................. 62
Table 3. Summary Statistics of Characteristics in VAR-STR Model .......................... 63
Table 4. Correlations between Characteristics in VAR-STR Model ........................... 64
Table 5. News Announcement Filter ........................................................................... 65
Table 6. Estimation Results of STR Model ................................................................. 66
Table 7. Estimation Results of News Surprise ............................................................ 67
Table 8. Estimation Results of Pure News .................................................................. 72
Table 9. Estimation Results of Good and Bad News .................................................. 76
Table 10. Robustness Results of Surprise on Alternative Slopes ................................ 77
Table 11. Robustness Results of Surprise on Alternative Volatilities .......................... 80
Table 12. Number of Significant News ....................................................................... 82
Table 13. Number of Significant News in Robustness ................................................ 83
Table 14. Robustness Results of Surprise on Depth at Ask and Bid Sides ................. 84
Table 15. Robustness Results of Surprise on Slope at Ask and Bid Sides .................. 87
Table 16. Robustness Results of Surprise on Volatility at different levels in the LOB
...................................................................................................................................... 88
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB .... 90
Table 18. Robustness Results of Surprise on Slope at different levels in the LOB .... 95
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
List of Figures
Figure 1. Intraday Pattern of Characteristics .............................................................. 99
Figure 2. Intraday Announcement Cluster ................................................................ 100
Figure 3. Transition Variable ISM ............................................................................. 101
Figure 4. Estimation Results of Logistic Transition Function .................................. 102
Figure 5. Intraday Patterns of Alternative Characteristics ........................................ 103
Figure 6. Autocorrelation Coefficients of Log Transformed Filtered Volatility ....... 104
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
1. Introduction
The effect of macroeconomic news on the dynamics of the limit order books
(LOB) has been investigated in previous studies. Erenburg and Lasser (2009) show
the release of the macro news affects the depth and spread of the LOB. They find that
macroeconomic announcements lead to deterioration in the quality of LOB. Andersen
et al. (2003) demonstrate that macro news announcements have a lasting effect on
exchange rate volatility.
The recent U.S. crisis in 2008 caused large fluctuations in LOB liquidity and
volatility in FX ECN market (Mancini et al., 2012). Laakkonen and Lanne (2010)
verify that news effects depend on economic states, and they find that bad news has a
stronger effect on exchange rate volatility during an economic expansion. Ben
Omrane and Savaser (2013) document that the impact of news varies over business
cycles. They show that nearly one third of the most important macro news has
sign-switching effects during the recent global crisis. Most previous studies that
examine news effects in different business phases, such as recession and expansion,
focus on the return or volatility of foreign exchange.
Instead of using the National Bureau of Economic Research (NBER) dates,
Laakkonen and Lanne (2010) use the Smooth Transition Regression (STR) model
(Teräsvirta, 1994) to measure business cycles. STR is a more accurate and detailed
method to identify the states of economy continuously by using Institute for Supply
Management Survey index (ISM) as a business regime indicator.
2
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Characteristics of the LOB describe liquidity and volatility. Previous studies point
out that limit orders play an important role in understanding the market structure of
the electronic trading systems. Usually, these studies use characteristics based on the
limited portions of LOB. Ahn et al. (2001) analyze the role of limit orders in the
liquidity provision in the stock market. They use depth and price volatility to illustrate
the dynamics between the order book state and order flow for the ask and bid sides.
However, recent studies verify the presence of information beyond the best quote
level in LOB. Cao et al. (2004) argue that the quotes beyond the best bid and ask of
the LOB contain important information.
Since previous literature focuses primarily on the news effects on asset return or
volatility during business cycles, with only limited information from the LOB, such as
the best quotes, we contribute by examining the response of four LOB characteristics
to macroeconomic news during different business cycles by using all LOB levels. We
investigate the effect of macroeconomic news on LOB dynamics by using
characteristics which can fully describe the shape of LOB. Besides volatility, we
choose spread, depth and slope to describe the liquidity of LOB. Coppejans et al.
(2001) and Naes and Skjeltorp (2006) find that these three characteristics are
correlated, while describe different LOB aspects.
Our data is the euro-dollar exchange rate and 89 news categories from the US and
Euro zone countries from Jan. 3rd 2006 to Dec. 1st 2009. To avoid noise present in
tick-by-tick data, we use 5 minutes intervals in our sample (Gunther, W. 2008). We
3
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
construct a vector auto regression (VAR) model to investigate the dynamics among
characteristics by using information from all price levels in the LOB. Furthermore, we
apply a STR on VAR to examine the response of characteristics to the
macroeconomic news in different business regimes.
Our results show that macro news has effects on LOB characteristics and the
news effects vary with the states of the economy (recession or expansion) and the
background of the crisis. Depth is more affected by the news surprise and pure news
during the expansion. Quoted spread and volatility are more affected by news surprise
during the expansion but have a stronger response to pure news during the recession.
Slope is more affected by the news surprise and pure news during the recession. For
good and bad news, depth, quoted spread and volatility (but not slope) show
significantly stronger response during the expansion. We find that the news related to
housing market and news viewed as a business indicator are consistently significant in
the recession or expansion for all four characteristics.
In summary, our study contributes to the existing literature in several ways. First,
we empirically show that macro news significantly affects LOB characteristics. Using
a VAR-STR model, our study shows that the LOB response of macro news varies
with different states of economy. Second, we use all LOB quote levels to construct the
LOB characteristics and find that, by considering full information, LOB
characteristics react more intensely to macro news, providing empirical evidence that
the upper levels in LOB are informative. The remainder of the thesis is organized as
4
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
follows: Section 2 reviews the literature. Section 3 presents the methodology. Section
4 describes the data. The results are discussed in the Section 5. Robustness check is
shown in the section 6. Section 7 concludes the findings.
2. Literature Review
Information reflected in LOB is currently of interest in a growing literature. Most
previous studies about LOB information use only the best quotes or a limited portion
of the LOB. Recent evidence indicates the presence of information beyond the best
quote level.
Most empirical studies find that LOB information is reflected in certain
characteristics of the book.
Biais et al. (1995) use up to five best quotes of LOB in
the Paris Bourse to study the LOB liquidity by using spread and slope to measure
supply and demand. They conclude that this information is useful when predicting the
liquidity of stock market. A growing body of literature discusses the relation between
public announcements and these LOB characteristics. Macroeconomic news affect the
exchange price directly or affect it indirectly by influencing the order flow in the LOB.
Nearly one-third of the news response contribute to the volatility of exchange rates
(Evans and Lyons, 2008; Love and Payne, 2008).
The range of the news has been expanded to describe different kinds of news
effects.
Riordan et al. (2013) study the effect of news on trading intensity, liquidity,
and volatility of stocks traded on the Toronto Stock Exchange (TSE). They categorize
5
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
news as “positive”, “negative” and “neutral” and find that the adverse selection costs
around the arrival of negative messages is higher than that of positive or neutral
messages. Bauwen, Ben Omrane and Giot (2005) categorize nine kinds of scheduled
and unscheduled news in the euro/dollar market. They address the influence of
scheduled and unscheduled news announcements in three phases: pre-announcement
periods, contemporaneous and post-announcement periods. They find that volatility
increases just before scheduled news releases.
Other characteristics are influenced by the macro news. Erenburg and Lasser
(2009) document the influence of spread, depth and volatility after scheduled macro
news releases by using the LOB data of the Island ECN. They find that spread
increases and depth decreases when the news occurs, which agrees with anecdotal
evidence that traders prefer to submit more limit orders when volatility is high. In
other words, traders tend to more be aggressive around news releases.
Another research area focuses on the effect of news on the LOB characteristics in
different economic phases. Andersen et al. (2003) find the occurrence of news
announcements triggers return variation. They show that bad news has greater
influence compared to the good news, which indicates an asymmetry effect, and that
bad news in “good times” have a larger impact compared to “bad times”. Laakkonen
and Lanne (2010) study effect of macro news on the volatility of EUR/USD exchange
rate over the states of the economy. By using the STR model with Institute for Supply
Management Survey (ISM) as transition variables (Teräsvirta, 1994), they capture the
6
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
state-dependent effect of news on volatility. Moreover, they sort the macroeconomic
news into two categories: good news and bad news. They find that bad news has a
stronger effect on volatility then good news. Ben Omrane and Savaser (2013)
document the effect of scheduled and unscheduled news on exchange rates from 2005
to 2009. They find the sign effect of some news will change in different business
cycles and investigate the factors that contribute to this sign-switching effect.
Recent empirical work suggests that connections exist among various LOB
characteristics, such as the interactions between the liquidity characteristics of spread,
depth and slope. The inside spread (difference between minimum ask price and
maximum bid price) is a key indicator of liquidity which increases if the spread
decreases. A very narrow spread indicates a liquid market; and depth is associated
with quoted and effective spreads, especially for heavily traded stocks (Bessembinder,
2002). Liquidity increases if depth increases. Larger depth indicates high degree of
trading intensity.
A deep market can be expected to absorb larger buy and sell orders.
Depth by Riordan et al. (2013) is computed from three price levels in the LOB. Slope
is another effective way to access the LOB’s information and it measures liquidity
intensity or the elasticity that responds to the change of demand and supply curves.
Generally, deep markets will have smaller bid-ask spreads because of the
increased competition among market makers for order flow (Cao et al., 2004).
Theoretically, Coppejans et al. (2001) develop a model of market trading and predict
an inverse relationship between depth and volatility.
Furthermore, as stressed by
7
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Duong and Kalev (2008), Naes and Skjeltorp (2006), slope is negatively related to
price volatility. Using a VAR model applied to a maximum of five quotes levels in
limit orders, Beltran, Durre and Giot (2004) study the ex-ante and ex-post
relationships between volatility and liquidity to discover that liquidity declines when
volatility rises, causing larger trading costs.
Since our study is based on highly frequency tick by tick intraday time series,
which is time stamped to million seconds, we should pay attention to the two issues.
The first problem is related to the formation of the LOB in which successive orders
are submitted at irregular times. One way to deal with irregular spaced data is to use
every event time update and tick record; the other is to use a joint time interval
(Bauwens and Giot, 2001). In the case of high frequency data most authors tend to use
equally spaced data for their study.
The second problem is that intraday seasonality exists in high frequency time
series. The evidence of intraday seasonality patterns of the return volatility is
pervasive. Empirical studies by Engle and Russell, 1998 and Bauwens and Giot, 2001
address the problem of removing intraday seasonality. Several methods can be
applied to control or remove the seasonality.
Beltran, Durre and Giot (2004) remove
the seasonality effect by using trading day dummies. Usually ARCH and GARCH
models are used in the volatility of low-frequency time series. For high frequency
data, the effectiveness of ARCH and GARCH models is controversial.
Another
method for controlling seasonality is the intraday average observation model (IAOM),
8
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
introduced by Ben Omrane and Bodt (2007) where a Flexible Fourier Form (FFF)
model is used to eliminate the intraday seasonality pattern on volatility (Andersen and
Bollerslev, 1998, Andersen et al., 2003, Laakkonen and Lanne, 2010).
Recent literature argues that the 2008 crisis has influenced the global economy
from financial markets to fundamental industries. This recent global recession is
considered the worst financial crisis since the Great Depression of the 1930s and its
effect on FX market is documented in literature. Melvin and Taylor (2009) provide an
overview of the important events of the recent global financial crisis and their
implications for exchange rates and market dynamics after 2007. They use the Global
Financial Stress Index (FSI) to measure the severity of the crisis.
A crisis leads to a significant decrease in liquidity. Fratzscher (2009) models the
time-varying effect of US shocks on exchange rates. He finds that FX reserves,
current account positions and financial exposure are important in explaining the
response of exchange rates to the financial crisis. He also finds that negative
US-specific macroeconomic shocks during the crisis have triggered a significant
strengthening of the US dollar. Goldstein and Kavajecz (2004) focus on the liquidity
provision at the New York Stock Exchange during the crisis.
They show that
liquidity is diluted on the day after the market crash as the order book exhibited large
spreads and poor depth. Engle et al. (2012) analyze the liquidity and volatility in the
U.S. treasury securities market around the U.S. crisis and the following
“flight-to-safety” periods. They document that treasury market depth declines sharply
9
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
during the crisis, accompanied by increased price volatility. In addition, volatility and
depth at the best quotes exhibit a negative relationship and this relation becomes more
persistent during the crisis. They find that the treasury market during recovery has
lower market depth, along with higher trading size and greater price uncertainty.
3. Methodology
In this section, we introduce the method used to compute LOB characteristics over
5-min mid quote returns from tick data. Then we introduce Vector Auto Regression
Model (VAR) with macro news as an exogenous variable to analyze the response of
characteristics to macro news. The two-regime STR model is then augmented to the
VAR model to analyze the effect of macroeconomic news corresponding to different
economic phases.
3.1 Interval Characteristics
Instead of using every update in the LOB (tick-by-tick data), an
equally-spaced-interval is chosen as basic-unit of observation in our study. To
illustrate this point, we compare the pros and cons of tick-by-tick data and the interval
data. Following Gunther, W. (2008) and Andersen et al. (2007), we can reduce “noise”
in the high frequency data while keeping the intraday seasonality of the data by using
equally spaced intervals. However, lower frequency may not reflect the characteristics
of high frequency data. But noise in tick-by-tick data may obscure any change in the
LOB which affects estimation. The actual LOB updates as long as there is a price or
10
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
size change from the previous tick. Hence, most updates hardly change the LOB. But
some information is lost with aggregation. Normally, a longer the time duration for an
interval implies more information is lost, with a lower level of noise. We use
5-minute interval to keep intraday pattern of the characteristics and to follow the
standard practice of previous reach.
In this section, we show how to construct as 5 minute interval characteristic from
the tick characteristics. We compute a time-weighted average of all tick
characteristics in that interval. The weight is the inverse of the duration between each
tick to the end point of the corresponding 5 minute interval (Bauwens et al., 2005). To
start with, we introduce the basic notation for a 5-min interval 𝑛 = 1,2, … , 𝑁, where
𝑁 = 288 the total number of intervals are in a sample day which is 24 hours, for
trading day 𝑡 = 1,2, … , 𝑇, where 𝑇 is the total number of days in the sample period.
Diagram 1 represents a typical tick in the LOB.
Following the method of Bauwens et al. (2005), we label the time duration
between each LOB update and the interval endpoint as 𝜏 seconds. 𝜏𝑖 represents the
time duration between 𝑖 𝑡ℎ tick in an interval n and the end time of the interval n,
where 𝑖 = 1,2, … , 𝛾𝑛 where 𝛾𝑛 is the total number of ticks in the interval n. So the
time duration in an interval n is 𝜏1 , 𝜏2 … , 𝜏𝛾𝑛 . For instance, as shown in Diagram 1, 𝜏1
is the time duration between the 1st tick in the interval n=2 and the end point of the
interval at n=2. The end point time is by definition. Let 𝑋𝑡,𝑛,𝑖 be a representative for
11
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
one of the characteristics in the LOB1. And the characteristic for interval n on day t is
defined as:
𝛾
𝑋𝑡,𝑛 =
1
𝜏𝑖
𝛾𝑛 1
∑𝑖=1
( )
𝜏𝑖
𝑛 [( )×𝑋
∑𝑖=1
𝑡,𝑛,𝑖 ]
,
(1)
where 𝑋𝑡,𝑛 is the time-weighted average of characteristics in interval n at trading
day t. 𝑋𝑡,𝑛,𝑖 is the characteristics in tick i, interval n at trading day t. In section 3.2,
characteristics of the LOB are defined based on tick-by-tick data and then the
characteristics for an interval n are calculated with equation (1).
In this case, the third tick which is updated nearest to the end of the 5-min
interval has the largest weight. Because the weight is the inverse of the time duration
between each tick updates to the end of its corresponding 5 minute interval. Since the
time duration of the update which is closest to the end of the interval is the shortest,
the tick that is nearest to the end of the interval has the largest weight.
In Appendix D, Diagram 1 shows an example of updates in the LOB. From the
diagram, the first interval has three ticks, 𝑖 = 1, 2, 3. 𝜏3 is the time duration of the
tick i=3 which is nearest to the end of the second interval on day t. According to the
equation above, the characteristics at tick i=3 have largest weight in forming that
interval’s characteristics.
1
𝑋𝑛,𝜏𝑖 are the depth, the quoted spread and the slope and mid-price which is used to define the volatility of the
LOB.
12
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
3.2 Characteristics of Limit Order Book
We measure the LOB information in two ways: liquidity and volatility. There are
three different measures of liquidity: the quoted spread, the depth, and the slope.
Spread reflects the magnitude of transaction cost and it measures how far the best ask
price is from the best bid price. Depth measures the amount of liquidity in the LOB.
And the slope measures the elasticity of the demand curve and supply curve of LOB.
The last characteristic is volatility which measures the fluctuation and transition cost
of LOB.
The Depth
The most cited depth method is simply the number of quoted forex units of each
corresponding price level. However, to make full use of the information in the LOB,
we follow Riordan et al. (2013) to compute the LOB depth. The robustness check for
alternative depth methods of depth is in section 6.
According to Diagram 1 in Appendix D, denote the price level at tick 𝑖 as 𝑙 =
1,2,3 … , 𝐿, where 𝐿 is the total number of the price levels in that tick i. The ask price
𝐴
at tick i with price level 𝑙 in interval 𝑛 is 𝑃𝑛,𝑖,𝑙
and the best ask price at each tick 𝑖
𝐴
𝐵
is 𝑃𝑛,𝑖,1
. The bid price at each tick 𝑖 with depth level 𝑙 in interval 𝑛 is 𝑃𝑛,𝑖,𝑙
. E.g.
𝐵
the best ask price at each tick i is 𝑃𝑛,𝑖,1
. Assume the size on a certain price level at the
𝐴
𝐵
ask side is 𝑄𝑛,𝑖,𝑙
, similarly the size on a certain price level at the bid side is 𝑄𝑛,𝑖,𝑙
.
𝐴
𝐵
Thus the size at the best ask level is 𝑄𝑛,𝑖,1
and the size at the best bid level is 𝑄𝑛,𝑖,1
,
𝐴
while the size at the second best ask level is 𝑄𝑛,𝑖,2
. According to Ryan Riordan et al.
13
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
(2013), the depth for a tick is the sum of price-weighted sizes of all levels in that tick.
Then the depth measure for every tick 𝑖 at interval n is 𝐷𝑒𝑝𝑡ℎ𝑛,𝑖 :
𝐴
𝐴
𝐵
𝐵
𝐷𝑒𝑝𝑡ℎ𝑛,𝑖 = ∑𝐿𝑙=1[𝑄𝑛,𝑖,𝑙
× 𝑃𝑛,𝑖,𝑙
] + ∑𝐿𝑙=1[𝑄𝑛,𝑖,𝑙
× 𝑃𝑛,𝑖,𝑙
],
(2)
𝐴
𝐵
where 𝑄𝑛,𝑖,𝑙
is the size of level l at tick 𝑖 in interval 𝑛 at ask side; 𝑄𝑛,𝑖,𝑙
is the size
of level l at tick 𝑖 in interval 𝑛 at bid side. The depth at interval 𝑛 in day 𝑡 is:
𝛾
𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 =
1
𝜏𝑖
𝑛 [( )×𝐷𝑒𝑝𝑡ℎ
∑𝑖=1
𝑡,𝑛,𝑖 ]
𝛾
1
𝜏𝑖
𝑛( )
∑𝑖=1
,
where 𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 is the time-weighted average of depth at interval n in day t. From
equation (2), depth by Riordan et al. (2013) is the sum of price-weighted size for both
sides of the LOB. Depth is sum of the price-weighted size for every price level l in
each tick i; in this case, depth is to measure the amount of liquidity in LOB. The depth
by Riordan et al. (2013) is more informative than simply using the size to describe the
amount of liquidity because the size here is weighted by the corresponding price at
each level.
The Spread
The quoted spread is a good indicator of the execution cost for a trade in case of
small orders. Also, the spread is influenced by market impact. For example, the spread
may be larger due to the size of the order (Riordan et al, 2013). In this case, we use
the quoted spread to measure the spread of LOB.
14
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
For every tick i in interval n, the quoted spread measured in the basis points is:
𝑄𝑆𝑝𝑟𝑒𝑎𝑑𝑛,𝑖 = (1
2
𝐵
𝑃𝐴
𝑛,𝑖,1 −𝑃𝑛,𝑖,1
(𝑃𝐴𝑛,𝑖,1 +𝑃𝐵𝑛,𝑖,1 )
) × 10000.
(3)
Then the construct the interval quoted spread by using the method in section 3.12.
The quoted spread at interval 𝑛 in day 𝑡 is,
𝛾
𝑄𝑆𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 =
1
𝜏𝑖
𝑛 [( )×𝑄𝑆𝑝𝑟𝑒𝑎𝑑
∑𝑖=1
𝑡,𝑛,𝑖 ]
𝛾
1
𝜏𝑖
𝑛( )
∑𝑖=1
,
where 𝑄𝑆𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 is the time-weighted average of quoted spread in interval n in day
t. Quoted spread is different from the minimum spread which is the difference
between the best ask price and the best bid price. From the equation (3), quoted
spread is more informative and easier to interpret because it is defined as the
percentage of the difference between best ask and bid quote to the mid-price, quoted
spread is a percentage measure of trade execution cost in for the mid-price of every
quote.
The Slope
Slope is a common information feature to measure the elasticity of the demand
and supply curves. Naes and Skjeltorp (2006) and Duong and Kalev (2008) use daily
data to measure the average slope across all price levels with LOB sizes considered.
They calculate the average of bid and ask slope to get one slope measure for each tick
by considering up to five price levels in a tick. They take the average across all the
ticks to obtain one daily average slope.
2
Multiply by 10,000 to enhance readability of the numbers multiplies the original quoted spread method. Scaling
quoted spread by 1,000,00 does not change its statistical properties (Riordan et al., 2013)
15
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
The interval slope uses the same procedure as the previous characteristics. The
𝐴
slope at ask side for each 𝑖 in an interval 𝑛 is 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
. Then we compute the
time-weighted average of the slope for ask side and get the interval slope at ask side:
𝐴
𝑆𝑙𝑜𝑝𝑒𝑡,𝑛
. For the ask side, the slope for each tick 𝑖 in an interval 𝑛 is
𝑞𝐴
1
𝐴
𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
= [𝑃𝐴 𝑛,𝑖,1 + ∑𝐿−1
𝑙=1 |
𝐿
𝑛,𝑖,1
−1
𝑚𝑛,𝑖
𝑞𝐴
𝑛,𝑖,𝑙+1
𝑞𝐴
𝑛,𝑖,𝑙
𝑃𝐴
𝑛,𝑖,𝑙+1
𝑃𝐴
𝑛,𝑖,𝑙
−1
|] ,
(4)
−1
𝐴
with l=1, 2, ...L as the price level with in each tick i. 𝑞𝑛,𝑖,𝑙+1
is the natural logarithm
𝐴
𝐴
ask size at tick level l+1 and 𝑞𝑛,𝑖,𝑙
is the ask size at tick level l. 𝑞𝑛,𝑖
is the natural
logarithm of the sum of the size at all levels in a particular tick i. For instance,
𝐴
𝐴
𝑞𝑛,𝑖,1
= ln(𝑄𝑛,𝑖,1
).
1
𝐴
𝐵
Denote mid-price as 𝑚𝑛,𝑖 = 2 (𝑃𝑛,𝑖,1
+ 𝑃𝑛,𝑖,1
) and denote
𝐴
𝑄𝑛,𝑖,0
= 0.
𝐵
For the bid side, 𝑞𝑛,𝑖,𝑙+1
is the natural logarithm bid size at tick level l+1 and
𝐵
𝑞𝑛,𝑖,𝑙
is the bid size at tick level l. we take the absolute value of each term in the
𝐵
equation of the slope of the ask side and get 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
with l=1, 2, ...L as the price
level in each tick i. The reason we take the absolute value of the slope of the bid side
is that best bid price is smaller than the mid-price, also we want to get the magnitude
of the elasticity. Lastly, taking the simple average of the ask slope and bid slope to get
the slope at interval 𝑛 in day 𝑡 at tick i is 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖 =
𝐴
𝐵
𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
+𝑆𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
2
.
16
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
The slope of the LOB at interval 𝑛 in day 𝑡 is,
𝛾
𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 =
1
𝜏𝑖
𝛾𝑛 1
∑𝑖=1
( )
𝜏𝑖
𝑛 [( )×𝑆𝑙𝑜𝑝𝑒
∑𝑖=1
𝑡,𝑛,𝑖 ]
,
where 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 s the time-weighted average of slope in an day t interval n. To
understand the equation (4): on the one hand, the first term in the bracket is the slope
from the midpoint to the best ask price level in a tick. In other words, the first term
measures the percentage of size at the best quote to the change of best ask price
relative to the mid-price.
The second term in the bracket is the sum of the slopes for
the rest levels in that tick. For each level, slope is the elasticity that measures the
change of sizes relative to the last level to the change of prices relative to the last level.
In summary, we measure the percentage change of size at every price level compared
to the size of the previous level in a tick. In other words, we measure the elasticity of
size with respect to in a tick.
Note that the first term and the second term are not measured in the same units.
(Naes & Skjeltorp, 2006). Since the size at the midpoint is unobtainable. We cannot
calculate the elasticity of the first term. In section 6, we summarize several different
measures of slope as a robustness test of slope measure.
So we have defined the slope, depth and quoted spread of the LOB and refer to
these three characteristics as “liquidity characteristics” where each measures the
liquidity of the LOB in different ways. Another reason to differentiate them from the
volatility is that the intraday seasonality patterns are different.
In other words, the
intraday seasonality patterns of liquidity characteristics are similar, but are different
17
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
from the volatility intraday seasonality pattern. In the last part of section 3.2, we
define the variation of the prices in LOB as the volatility for the LOB return. In 3.2.4,
we define the mid-price of LOB then define the return volatility.
The Volatility
Kozhan and Salmon (2010) demonstrate the economic value of LOB information
in FX markets by using the full book. In this paper, they use size-weighted price to
calculate the mid-price and spread. In other words, the mid-price and spread combines
all the levels of the LOB.
We follow Kozhan and Salmon (2010), and compute the
size-weighted average price of LOB. Compared to the methods that use the best quote
price to calculate the return, the method by Kozhan and Salmon (2010) combine the
size and prices of all levels in one tick.
The average ask price at tick 𝑖 in interval n is
𝐴𝑃 𝐴 𝑛,𝑖 =
𝑨
𝐴
∑𝐿𝑙=1[𝑄𝑛,𝑖,𝑙
×𝑃𝑛,𝑖,𝑙
]
𝑨
∑𝐿𝑙=1 𝑄𝑛,𝑖,𝑙
,
(5)
where 𝐴𝑃 𝐴 𝑛,𝑖 is the size-weighted average ask price at tick i in interval n and
𝑨
denote 𝑄𝑛,𝑖,0
= 0. Likewise, we denote 𝐴𝑃𝐵 𝑛,𝑖 as the average bid price at tick i in
interval n.
The size-weighted price means that the price at each level is weighted by
the percentage of the corresponding size on that level to the total size of all levels in
that tick. Denote 𝑀𝐼𝐷𝑛,𝑖 is the average mid quote price at tick i in interval n:
𝐴𝑃 𝐴 𝑛,𝑖 +𝐴𝑃𝐵 𝑛,𝑖
𝑀𝐼𝐷𝑛,𝑖 =
2
.
18
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
The mid-quote price at tick i in interval n is:
𝛾
𝑀𝐼𝐷𝑡,𝑛 =
1
𝑛 [( )×𝑀𝐼𝐷
∑𝑖=1
𝑡,𝑛,𝑖 ]
𝜏
𝑖
𝛾
1
𝜏𝑖
𝑛( )
∑𝑖=1
,
where 𝑀𝐼𝐷𝑡,𝑛 is the time-weighted average of mid-quote prices in interval n on day t.
We calculate the return based on the mid quote.
Following Andersen and Bollerslev
(1998), the return in interval n at sample day t is: 𝑅𝑡,𝑛 = (𝑙𝑜𝑔(𝑀𝐼𝐷𝑡,𝑛+1 ) −
𝑙𝑜𝑔(𝑀𝐼𝐷𝑡,𝑛 )) × 100, where 𝑅𝑡,𝑛 is the return over the 5-min interval n=1,2,…, N
for sample day t=1,2,3,… T. Then we obtain the absolute centered 5-min return
structure|𝑅𝑡,𝑛 − 𝑅̅ |, denoted here as Abs_return, where 𝑅̅ is the average return for
whole sample.
3.3 VAR with Two-regime Smooth Transition Regression
Using data of Dow Jones (DJIA) stocks, Nigmatullin, Tyurin, and Yin (2007)
show that the significant interactions exist among characteristics of LOB in a Vector
Auto Regression (VAR) model. Following Nigmatullin, Tyurin, and Yin (2007), we
construct a model to analyze the effect of macro news on the characteristics in the
LOB in business cycles. In other words, we use a VAR model to describe the joint
dynamics among the characteristics with macroeconomic news being exogenous
variables. To be specific, we construct a VAR model with j-lagged endogenous
variables, VAR (j).
In the section 3.3.1, STR is introduced into VAR for measuring the effect of
macro news on characteristics in different economic cycles. In the section 3.3.2, we
19
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
construct a VAR-STR model to estimate the effects of the news surprise in different
economic cycles. The estimation methods of the effect of pure news and aggregated
good and bad news are introduced in section 3.3.3 and 3.3.4.
Two-regime Smooth Transition Regression
We follow Laakkonen and Lanne (2010) to detect the regime transition by
applying the two-regime logistic smooth transition regression (LSTR) (Teräsvirta,
1994). The LSTR is:
𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 = 𝛼1 + ∑𝐽𝑗=1 𝛽𝑗 𝛤𝑡,𝑛−𝑗 + {𝛼2 + ∑𝐽𝑗=1 𝛽𝑗′ 𝛤𝑡,𝑛−𝑗 }𝐺(𝜓𝑡,𝑛 , 𝛾, 𝑐) + 𝜀𝑡,𝑛 ,
1
with 𝐺(𝜓𝑡,𝑛 , 𝛾, 𝑐) = 1+𝐸𝑋𝑃[−𝛾 ∏𝐾
𝑘=1(𝜓𝑡,𝑛 −𝑐𝑘 )]
, 𝛶 > 0.
(6)
(7)
Denote 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 as the log-transformed filtered volatility on day t and interval n
after the filter for intraday seasonality effects and daily ARCH effects; and 𝛤𝑡,𝑛−𝑗
includes consolidated macroeconomic news. The common choice of k is either one or
two. If k=1, this is a logistic STR1 model. Transition function 𝐺(𝜓𝑡,𝑛 , 𝛾, 𝑐) is a
logistic function of the continuous transition variable 𝜓𝑡,𝑛 . The transition variable is
represented by ISM index figures 3 . The model implies transition between two
economic regimes: higher regime (𝐺(𝜓𝑡,𝑛 , 𝛾, 𝑐)=1 when 𝜓𝑡,𝑛 > 𝑐), and lower regime
(𝐺(𝜓𝑡,𝑛 , 𝛾, 𝑐)=0 when 𝜓𝑡,𝑛 < 𝑐), where 𝛾 is the shape parameter, c is the location
parameter, and k is the transition function scale. If the shape parameter 𝛾 is high, this
indicates a sudden transition happened during the sample period.
3
To estimate the regime transition for U.S. crisis, we choose the corresponding transition variable based on the
ISM (Institute for Supply Management) manufacturing index for US business cycles.
20
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
VAR-STR for Surprise
First we model the characteristics’ response to the news surprise based on the
VAR model with exogenous variables. Surprise measures the magnitude of news
effect. In this case, there are three types of exogenous variable in the model, the news
surprise of macroeconomic news, and a seasonality dummy for liquidity
characteristics.
Then, after the identification of economic regimes in the sample, we introduce the
US crisis in a VAR model by imposing the fitted logistic transition function
𝐺̂ (𝜓𝑡,𝑛 , 𝛾, 𝑐).
Then we combine the VAR and STR models in the case of a news
surprise. With 𝑙 lagged values in the characteristic variables, the VAR-STR model in
matrix notation is:
𝑈𝑆
Ω𝑡,𝑛 = 𝛼𝑡,𝑛 + ∑𝐽𝑗 𝛽𝑗 Ω𝑡,𝑛−𝑗 + 𝜆𝐴𝑉𝑡,𝑛 + ∑𝑄𝑞=1 𝜃𝑞 𝑆𝑞,𝑡,𝑛 + 𝜂1 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶
𝑈𝑆
′
𝜂2 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+ {𝛼𝑡,𝑛
+ ∑𝑄𝑞=1 𝜃𝑞′ 𝑆𝑞,𝑡,𝑛 + 𝜂1′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶 ̂
𝜂2′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
}𝐺 (𝜓𝑡,𝑛 , 𝛾, 𝑐) + 𝜀𝑡,𝑛
(8)
where characteristics Ω𝑡,𝑛 is a vector of endogenous variable which represents one of
four characteristics as the vector of dependent variables. Hence the vector of
′
endogenous variables in (8) is: Ω𝑡,𝑛 = (𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 , 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 , 𝑄𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 , 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 ) ,
where 𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 is the depth at interval n on day t; 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 is the slope at interval n
on day t; 𝑄𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 is the quoted spread at interval n on day t; and 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 is the
′
filtered volatility at interval n on day t. 𝛼𝑡,𝑛 and 𝛼𝑡,𝑛
are vectors of the constant
(intercepts). 𝜀𝑡,𝑛 is the error term. 𝛽𝑗 is coefficient matrix of Ω𝑡,𝑛−𝑗 and 𝜆 is the
21
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
coefficient matrix of vectors of seasonality dummies. Lag 𝑗 = 1,2, … , 𝐽4 is decided
by AIC and BIC criteria.
When liquidity characteristics (depth, slope or quoted spread) are dependent
variables respectively, each dependent variable has its corresponding seasonality
dummy as an exogenous variable to control the effect of intraday seasonality in
estimation. We denote the intraday seasonality dummy for liquidity characteristics as
𝐴𝑉𝑡,𝑛 which represents one of the vectors of the seasonality dummy of liquidity
characteristics: quoted spread, depth and slope respectively.
seasonality
dummy
of
liquidity
characteristics
The vector of the
is:
𝐴𝑉𝑡,𝑛 =
′
𝑑𝑒𝑝𝑡ℎ
𝑞𝑠𝑝𝑟𝑒𝑎𝑑
𝑠𝑙𝑜𝑝𝑒
(𝐴𝑉𝑡,𝑛
, 𝐴𝑉𝑡,𝑛
, 𝐴𝑉𝑡,𝑛
) . The seasonality dummy of quoted spread, depth and
𝑑𝑒𝑝𝑡ℎ
𝑞𝑠𝑝𝑟𝑒𝑎𝑑
𝑠𝑙𝑜𝑝𝑒
slope are 𝐴𝑉𝑡,𝑛
, 𝐴𝑉𝑡,𝑛
or 𝐴𝑉𝑡,𝑛
respectively. Note that 𝐴𝑉𝑡,𝑛 is a
regressor when liquidity characteristics are endogenous variables only. We use the
IAOM method to construct the “seasonality dummy” of three liquidity characteristics.
For volatility, we use FFF method to filter out the intraday seasonality in volatility.
The filtered method of liquidity characteristics and volatility is also in section 3.4.
Denote the news categories 𝑞 = 1,2,3, … , 𝑄, where 𝑞 indicates the one of
categories of macroeconomic news and 𝑄 is the total number of macroeconomic
news announcements in the sample. 𝐺̂ (𝜓𝑡,𝑛 , 𝛾, 𝑐) is the fitted value of logistic
transition variable in LSTR. Following the method by Balduzzi Elton and Green
(2001), we denote 𝑆𝑞,𝑡,𝑛 as the news surprise for news category q in interval n on
4
Using AIC and BIC criteria, the results indicate that a 1-lag structure (j=1) is adequate we estimate a VAR(1)
with four variables.
22
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
day t. The calculation of news surprise is shown in section 4. The surprise vector of
coefficients is denoted as 𝜃𝑞 ; the transition vector of surprise coefficients is denoted
as 𝜃𝑞′ . We also consider unscheduled new related to the US crisis as the other two
exogenous variables. One is unscheduled news related to crisis for the US and the
𝑈𝑆
other is unscheduled news related to the European crisis. 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
is the US
𝐸𝐶
unscheduled news and 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
is the European unscheduled news. The vector of
coefficients of unscheduled news related to the US and EC crisis are denoted as 𝜂1
and 𝜂2 , respectively; the other two coefficients vector of unscheduled news with
respect to the
transition variable are denoted as 𝜂1′ and 𝜂2′ respectively.
VAR-STR for Pure News
We construct the STR model to examine the effects of pure news on
characteristics in different regimes. Pure news is different from the news surprise and
the VAR-STR model with pure news as exogenous variables is:
𝑈𝑆
Ω𝑡,𝑛 = 𝛼𝑡,𝑛 + ∑𝐽𝑗 𝛽𝑗 Ω𝑡,𝑛−𝑗 + 𝜆𝐴𝑉𝑡,𝑛 + ∑𝑄𝑞=1 𝜉𝑞 𝑃𝑢𝑟𝑒𝑞,𝑡,𝑛 + 𝜂1 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶
𝑈𝑆
′
𝜂2 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+ {𝛼𝑡,𝑛
+ ∑𝑄𝑞=1 𝜉𝑞′ 𝑃𝑢𝑟𝑒𝑞,𝑡,𝑛 + 𝜂1′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶 ̂
𝜂2′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
}𝐺 (𝜓𝑡,𝑛 , 𝛾, 𝑐) + 𝜀𝑡,𝑛
(9)
where 𝑃𝑢𝑟𝑒𝑞,𝑡,𝑛 denotes the pure news of category q in interval n at day t. 𝜉𝑞 is the
coefficient vectors of pure news of category q at date t and interval n, the other vector
of pure news coefficient with the effect of the transition variable is denoted as 𝜉𝑞′ .
23
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
VAR-STR for Aggregated Good and Bad News
We construct aggregated dummy for “good” news and “bad” news. So the
VAR-STR model with aggregated “good” and “bad” news as exogenous variables is:
𝑈𝑆
Ω𝑡,𝑛 = 𝛼𝑡,𝑛 + ∑𝐽𝑗 𝛽𝑗 Ω𝑡,𝑛−𝑗 + 𝜆𝐴𝑉𝑡,𝑛 + 𝜌𝑔 𝐺𝑜𝑜𝑑𝑡,𝑛 + 𝜌𝑏 𝐵𝑎𝑑𝑡,𝑛 + 𝜂1 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶
𝑈𝑆
′
𝜂2 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+ {𝛼𝑡,𝑛
+ 𝜌𝑔′ 𝐺𝑜𝑜𝑑𝑡,𝑛 + 𝜌𝑏′ 𝐵𝑎𝑑𝑡,𝑛 + 𝜂1′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝐸𝐶 ̂
𝜂2′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
}𝐺 (𝜓𝑡,𝑛 , 𝛾, 𝑐) + 𝜀𝑡,𝑛
(10)
where 𝐺𝑜𝑜𝑑𝑡,𝑛 denotes the aggregated “good” news in interval n at day t. 𝐵𝑎𝑑𝑡,𝑛
denotes the aggregated “bad” news in interval n at day t. 𝜌𝑔 and 𝜌𝑏 are the
coefficients vector of aggregated “good” and “bad” news in interval n on day t.
3.4 Intraday Seasonality
In this section, we introduce filter methods for the intraday seasonality pattern.
For liquidity characteristics (slope, depth and quoted spread), we use the intraday
average observation model (IAOM) to construct control dummies for seasonality
(Omrane and Bodt, 2007).
For volatility, we use Flexible Fourier Form (FFF) to
filter the seasonality (Anderson et al., 2003).
Intraday Seasonality Patterns of Liquidity Characteristics
For the spread, depth and slope, we adjust each variable for its intraday
seasonality by using the intra-day average observations model (IAOM) (Omrane and
Bodt 2007). The control dummies for three characteristics become exogenous
variables in the VAR to capture the intraday periodicity of those liquidity
characteristics. We remove all the intervals that belong to the following dates:
24
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Monday, Tuesday, Wednesday, Thursday and Friday, and exclude non-trading dates:
Saturday and Sunday. So we get five sub-sets of the sample data, one for each
weekday. Then calculate the simple average of corresponding characteristics for each
subset to get intra-day average observation for each interval in a trading day. Finally,
we construct the control dummy with the value of the intra-day average for all the
intervals in the five weekdays respectively. The corresponding exogenous control
dummies for endogenous variables 𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 , 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 and 𝑄𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 , are quoted
𝑑𝑒𝑝𝑡ℎ
𝑞𝑠𝑝𝑟𝑒𝑎𝑑
𝑠𝑙𝑜𝑝𝑒
spread (𝐴𝑉𝑡,𝑛
), depth (𝐴𝑉𝑡,𝑛
), and slope (𝐴𝑉𝑡,𝑛
).
Intraday Pattern of Volatility
The return volatility is also strongly correlated to market activities as Andersen &
Bollerslev (1998) conclude that the return volatility is affected by intraday activity
patterns. So we adjust our model to eliminate the influence of intraday seasonality
patterns.
Following Andersen & Bollerslev (1998), we decompose the volatility as:|𝑅𝑡,𝑛 −
𝑅̅ | =
𝜎𝑡
√𝑁
where 𝑅̅ is the sample mean return of 𝑅𝑡,𝑛 .
𝜎𝑡
√𝑁
represents a daily ARCH
effect, 𝜎𝑡 denotes the AR (2)-GARCH (1, 1) one day ahead daily volatility5, N is the
total number of 5-min intervals per day. Then square and take natural log on both
sides to obtain 2 ln (
|𝑅𝑡,𝑛 −𝑅̅ |
𝜎𝑡
√𝑁
) = 2ln(ℎ𝑡,𝑛 ) + 2ln(𝑣𝑡,𝑛 ) , where ℎ𝑡,𝑛 denotes the
intraday seasonality and 𝑣𝑡,𝑛 contains the rest of the volatility including
5
Appendix B shows the computation method of one day ahead volatility.
25
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
announcement effects. Next, we estimate the cyclical volatility component and use
Flexible Fourier Form (FFF) regression to get 𝑓̂𝑡,𝑛 :
𝑛2
𝑛
2𝜋𝑝
𝑃
𝑓𝑡,𝑛 = 𝜇 + 𝛿1 𝑁 + 𝛿2 𝑁 + ∑𝐷
𝑑=1 𝜆𝑑 𝐼𝑑 (𝑡, 𝑛) + ∑𝑝=1 (𝛿𝑐,𝑝 cos (
1
𝛿𝑠,𝑝 sin (
2
2𝜋𝑝
𝑁
𝑁
𝑛)) + 𝜀𝑡,𝑛
𝑛) +
(11)
|𝑅𝑡,𝑛 −𝑅̅ |
where 𝑓𝑡,𝑛 is the log-transformed volatility and 𝑓𝑡,𝑛 = 2 𝑙𝑛 (
𝑛
𝑁1
); 𝜇 is constant;
𝑛2
and 𝑁 are normalizing factors, here 𝑛 is the number of interval where 𝑁1 =
𝑁+1
2
𝜎𝑡 ⁄√𝑁
2
and 𝑁2 =
(𝑁+1)(𝑁+2)
6
. Normalizing factors are used to control for holiday
effects, weekday effects etc. 𝑅̅ is the expected intraday returns for size-weighted
mid-price. In addition, 𝜎𝑡 denotes one day ahead daily volatility in a GARCH (1, 1)
model using the interval return 6 . 𝐼𝑘 (𝑡, 𝑛) is an indicator for the event 𝑑 during
interval 𝑛 on day 𝑡. 𝐼𝑘 (𝑡, 𝑛) captures the calendar effects: Japanese open, Japanese
lunch and the U.S. late afternoon during U.S. daylight saving time. For Japanese
open events, a polynomial structure with the single order for 2 hours is used to
capture the increased log-volatility when Japan opens; and a second order polynomial
structure is applied to capture the volatility decay pattern for the summer regime. The
sinusoids denotes the Flexible Fourier Form that provides the approximation of the
intraday periodicity pattern. Choose p according to Schwarz and Akaike Information
Criteria 7 . In order to capture the deterministic and time varying seasonality
components, the FFF estimation is done in sequential sub periods of four weeks.8
6
The estimate of one-day-ahead volatility is forecasted based on the daily volatility from January 11, 2004 through December
31 2009.
7
8
According AIC, p=1;
We considered sub periods of one and two weeks but estimated results were not statistically significant.
26
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
The estimate of the normalized intraday seasonality is computed as 𝑠̂𝑡,𝑛 =
̂ 𝑡,𝑛
𝑓
)
2
exp(
𝑠̅𝑡,𝑛
, where 𝑓̂𝑡,𝑛 are the fitted values of the model. This estimate 𝑠̂𝑡,𝑛 is
normalized so that the mean of the normalized seasonality estimate equals one: 𝑠̅𝑡,𝑛 =
𝑇×𝑠̂𝑡,𝑛𝑘
𝑇/𝑁
∑𝑡=1 ∑𝑁
𝑛=1 𝑠̂𝑡,𝑛
, where T is the number of observations in the whole sample. Following
𝑘
Andersen and Bollerslev (1998), the original volatility 𝑅𝑡,𝑛 is then divided by the
𝑅
normalized estimate 𝑠̅𝑡,𝑛 to compute filtered returns: 𝑅̂𝑡,𝑛 = 𝑠̅ 𝑡,𝑛. Finally, the filtered
𝑡,𝑛
volatility for VAR regression is: 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 = 2𝑙𝑛
|𝑅̂𝑡,𝑛 −𝑅̅ |
𝜎𝑡 /√𝑁
, where 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 is filtered
volatility based on the return calculated by size-weighted mid-price of the LOB.
4. Data
4.1 Limit Order Book
The original book is obtained from Hotspot FXi. The book contains quote
tick-by-tick data from Jan 3rd 2006 to Dec 31st 2009.
The original LOB records limit
orders of exchange rates and size for Euro/Dollars.
Each tick in the LOB is stamped
as milliseconds in Eastern Standard Time (EST) adjusted daylight saving time.
Diagram 1 shows a simple example of a single update in the LOB. For example, four
levels at the ask side and three levels at the bid side of the first tick (i=1) in interval
n=2. For each level in the LOB, the price of that level and the corresponding size is
given. The best ask is the lowest ask price which is denoted as ask price level l=1 and
similarly the best bid is the highest bid price, denoted by bid price level l=1.
27
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
To demonstrate the LOB liquidity condition in different phases of economy, we
choose up to two months of the LOB before and during the crisis according to Figure
4 which shows the estimated economic transition regimes for the sample period. One
is January of 2006 (Panel A of Table 1), the other is April of 2009 (Panel B of Table
1). Table 1 shows the descriptive statistics of ask and bid side of the LOB in the
different stages of the business cycles. To measure the intensity of trading activity, we
calculate the number of levels in a tick and the number of ticks in a 5-min interval.
In January 2006, there are about 340 ticks per interval for both ask and bid sides.
In
the most intensive hour, the number of ticks can soar to 1134 per 5 minute interval,
while the most inactive hours has only one tick. These facts indicate the disparity of
the trading intensity of the FX market in a day.
Compared to the situation of January
2006, the LOB is illiquid during the crisis (April 2009). The average number of the
ticks in a 5-min interval is 289 with a maximum of 572 for both the ask and bid sides.
To measure the “deepness” of the book, the number of levels in a tick is listed in the
table. For January 2006, the mean number of tick level for both sides is 9 with a
maximum of 22, compared to 23 in April 2009.
The maximum number of level in a
tick for the ask side is 45 and compared to 49 for the bid side, which is around twice
the number in January 2006.
Size is another aspect to measure the LOB liquidity situation. Table 1 shows the
size of the best quote and the size of the whole book in one tick. Generally speaking,
the size in one tick before the crisis (January 2006) is much larger than the size of a
28
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
tick during the crisis (April 2009).
YUSI TAO
In the ask side, the size in a tick in Jan 2006 is
75900682 with an average level of 9 which is larger than in April 2009
(66115081with average level of 23).
Also, the size is more centrally distributed near
the best quote before the crisis compared to that during the crisis, which indicates a
steeper size curve. The best size on ask side is 17% of the total size in the LOB for
January 2006, compared to 5% on the ask side in April 2009. Table 1 also list the
spreads for both months. The spread is the difference between the best ask and bid
prices.
The average of the spread in the 2006 (0.00018) is a bit larger than that in
April 2009 (0.00016).
4.2 Interval Data
We resample the data with equally-divided-intervals instead of using tick-by-tick
data and choose 5-min intervals as a compromise between information and noise.
Since each quote is time stamped to milliseconds, the first interval of a trading day is
00:00:00.000 EST to 00:04:59.999 EST. Hence for one trading day, the time goes
from 00:00:00.000 EST to 23:59:59.999 EST. The sample data only includes trading
days (no weekends) and we exclude the outlier interval from 5:00 PM to 5:40 PM as
well as ten important US statutory holidays9. The first interval (00:00:00.000 EST to
00:04:59.999 EST) is deleted to avoid overnight effect. The data sets for estimation
include intraday 5-min interval euro/dollar exchange rate data and macroeconomic
news data from Jan 3rd 2006 to Dec 31st 2009.
The Ten US statutory holidays deleted are: New Year’s Day, Martin Luther King’s Day, Presidents Day,
Memorial Day, Independence Day, Labor Day, Columbus Day, Veterans Day, Thanksgiving Day and Christmas
Day.
9
29
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 2 provides the summary statistics for characteristics in an interval.
Excluding the outliers and important holidays, we have 245,780 intervals in sample
years from 2006 to 2009.
Note that these characteristics are not adjusted for intraday
patterns. The volatility based on the size-weighted average price is denoted as
“Abs_return”. The average of the volatility is 0.02% with a maximum 11.41%. The
other method introduced in section 6 is denoted as “Abs_ret”, which has an average of
0.03% with a maximum of 8.08%. Table 2 also lists the descriptive statistics of two
depth measures (Depth and size), two spread measures (Qspread, sizespread) and
three slope measures (slope, NORM SLOPE, WSLOPE). We show the
autocorrelation of each characteristic up to 2 lags. The null hypothesis that no
autocorrelation exists is rejected and the autocorrelation coefficients of all the
characteristics are significant at 2 lags.
Table 3 shows the summary statistics of standardized characteristics and Table 4
presents the correlations between standardized characteristics used in VAR-STR
regression. We standardized the liquidity characteristics (depth, size, slope,
NORMSLOPE, WSLOPE, quoted spread and size-weighted spread) with their
corresponding standard deviation. Focusing on the four characteristics we introduced
in section 3, we find depth is negatively related to the quoted spread, slope and
volatility, which agrees with the Ahn et al (2001). As shown in the Naes and Skjeltorp
(2006), volatility is negatively related to the slope. The correlation between quoted
spread and slope is negative, while quoted spread is positively related to volatility. In
30
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
addition, the t-test is conducted on these correlations with the null hypothesis that the
correlation between variables are zero which are all significant.
4.3 Macroeconomic News
The macroeconomic news data include scheduled and unscheduled news. There
are 89 categories of news announcements in different countries and regions in our
sample from 2006 to 2009.
Scheduled News
Supported by Bloomberg, the scheduled macroeconomic news includes the news
announcements from US, Euro Zone, Germany, France, Italy, Spain, and Poland.
Usually, news is released weekly, monthly and quarterly. We exclude news with very
few observations or missing actual or forecast values.
a.
News surprise
Bloomberg provides both actual and market forecasts of news announcements.
The market forecasts are the median value of the survey, which is conducted before
the release day. We consider both actual figures and forecasts by using the news
surprise, which is measured in Balduzzi, Elton and Green (2001). The news surprise
is the difference between the forecast and the actual figure and then dividing this
difference by the standard deviation of these differences:
𝑆𝑞,𝑛 =
𝐴𝑞,𝑛 −𝐹𝑞,𝑛
̂𝑞
𝜎
where 𝐴𝑞,𝑛 the actual figures of news q at interval n is, and 𝐹𝑞,𝑛 is the forecasts for
the corresponding announcements. 𝜎̂𝑛 is the sample standard deviation of the
31
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
difference between actual and forecast for news q considering in the sample.
The
news surprise measures the response of the market to the news.
b. Pure News
To measure the effect of a news occurrence, a “pure news” dummy is created for
every category of scheduled macroeconomic news with different countries. The
dummy variable is one when news happens, otherwise it is zero. In all, we have 3452
news surprises of news announcements and 89 news categories. The news surprise is
0 when the actual figure of news equals to the forecast for the news. Even if news
surprise is zero, a news event still happens. In 3452 news surprises, 444 have news
surprise of 0. In other words, we lose 13% of news in the estimation of the VAR-STR
model. So pure news can make up the 13% loss from the news surprise. The surprise
news can measure the magnitude of news effect, while pure news only catches the
event itself.
c.
Aggregated News
Besides the news surprise and pure news, aggregated news is applied in the
VAR-STR model to show the aggregated effect of news on the LOB characteristics.
Under the assumption that news effects can be aggregated, we construct a dummy for
aggregated news. We classify the news as “good” and “bad” for each category of
news announcement. “Good” news has a positive effect (appreciation) on currency.
“Bad” news has a negative effect (depreciation) on currency. We adjust some news
according to the meaning of the news, for example, when the unemployment rate is
32
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
underestimated, this indicates “bad” news. Then we aggregate all the “good” news
into a dummy of 1 for “good” news and 0 otherwise. Similarly, we construct the
dummy of aggregated “bad” news: 1 for “bad” news and 0 otherwise.
d. Filter Rules
With missing observations excluded in the first round of filters, we also have to
delete some news during the estimation process of VAR-STR. During the analysis,
we have to filter the news that leads to multicollinearity issues in regression equations.
Usually, the news that is deleted in this step has a linear correlation with other
variables in the regression. The news that has a significant coefficient in the
contemporaneous regression (Andersen et al., 2003) is kept.
Based on the estimation process, Table 5 summaries the number of news after
filtering.
In all, we have 89 categories of news. In the case of VAR-STR estimation
on the news surprise, only one news in the EC is deleted.
For the case of pure news,
the news dummy caused more singular matrix issues and 15 news are deleted. As
shown in the Table 5, with 3452 observation of the news announcements, only 2299
(67%) are applied in the estimation. In the case of regression of pure news, 67% of 89
news categories are used.
Unscheduled News
Previous literature points out that unscheduled news effects should be controlled
in the forex market (Bauwen, Ben Omrane & Giot, 2005). Unscheduled News related
to the Crisis is constructed based on the dates of the US crisis from Federal Reserve
33
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Bank of New York. The time line mark the days for important news related to a crises.
A dummy series was constructed with one for days with important news and zero
otherwise.
5. Empirical Results
In this section, first we analyze the intraday seasonality pattern of the
characteristics. Then we show the estimation results of VAR-STR model to surprise,
pure news and good/ bad news.
5.1 Characteristics Analysis
LOB Characteristics have intraday seasonality patterns and time periodicity
reflects the fact that the FX market trade in different time zones around the world
during a day. In general, there are two kinds of patterns of figures. One is the pattern
of liquidity characteristics, while the other is the volatility seasonality pattern. The
detailed method dealing with this seasonality is given in the section 3.
The intraday pattern of the four characteristics is plotted in Figure 1 using the
average for every interval during the sample. Figure 2 shows the cluster of news in a
day for euro zone countries10 and the US. Depth gradually increase often from the
midnight, that is, 00:05 EST, but stays at a relatively low level when the London and
New York markets are closed. Then the depth curve goes up after 2:00 EST when
several European markets are open. When the New York Market opens, the depth
10
The news euro countries: EC, FR, GE, IT, PO and SP.
34
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
reaches a higher level with two downward plunges around 8:00 and 10:00 EST which
agrees with the announcement times of most news. Then depth gradually declines to
reach its lowest level just after 17:00 EST when the New York market closes. After
17:00 EST, only a few markets in Asia and Australia are open. Although the activity
gradually increase after five, it stays lower than the amount during the daytime in
EST.
The intraday pattern for the quoted spread (Figure 1.b) experiences on opposite
tendency compared to depth. The quoted spread gradually goes down, and stays in a
relatively low when London and New York market are closed. When the majority of
US news announcements happen around 8:00:00 EST and 10:00:00 EST giving the
quoted spread two sudden peaks. After that, the European Markets start to close and
the quoted spread gradually increases to its peak right after New York market closes.
The slope intraday pattern tends to be more stable (Figure 1.c). Around 2:00 EST,
the slope gradually goes up and stays a high level with downward fluctuations around
8:00 EST and 10:00 EST, which considers with the majority of news announcements.
Then the slope gradually reaches its peak around 17:00:00 EST when New York
closes, indicating that the slope reaches its highest level of the day when the trader
aggressiveness is high. After 17:00:00 EST, the slope decreases dramatically and
stays relatively low.
The intraday pattern of volatility is shown in Figure 1.d. In addition, the volatility
quickly goes up, and reaches its first peak around 3:00 EST when the London market
35
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
opens. After that, volatility peaks around 8:00 EST and stays at a high level until
11:00 EST. Then the volatility experiences a declining undulation to reach its lowest
point at time when New York closes.
In section 3, we introduced two methods for controlling or eliminating these
intraday patterns. In the case of liquidity characteristics, we use IAOM to construct
average control dummies which are used in the estimation to capture the intraday
pattern of depth, slope and quoted spread.
In the case of volatility, we use FFF
regression to eliminate the intraday pattern in the volatility (Andersen & Bollerslev,
1998).
In addition to the volatility introduced in the section 3, Abs_return, we do a
robust check by using the volatility based on the best quote, Abs_ret (section 6.1.4).
FFF regression is applied in both two volatility measures.
Figure 6 plots the autocorrelation coefficients of filtered and original volatility.
To present the periodicity pattern in the auto correlogram, we plot the autocorrelation
coefficients of the original volatility in 5 days 11. As shown in panel A in Figure 6, the
autocorrelation coefficients of Abs_return shows a regular rising and falling after
going through a more intense fluctuation in its starting phase while in panel B, the
autocorrelation coefficients of Abs_ret before filtering moves in a regular wave.
Although filtered volatility is still auto correlated, the filtered volatility for both
Abs_return and Abs_ret moves stably around 0, implying that the FFF regression
eliminates the intraday seasonality pattern of the Abs_return and Abs_ret.
11
We calculate the autocorrelation of the original volatility for 1400 lags.
36
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
5.2 Estimation Results of the Logistic Transition Function in STR model
According to Veredas (2006), the ISM index is used as business cycles indicators
for regime transition effects and is more informative and accurate than NBER in
Laakkonen and Lanne (2010). ISM higher than 50 indicates that economy is in
expansion (good times). The ISM plot for 2006 to 2009 is in Figure 3. The ISM falls
below 50 between the fall of 2008 to the fall of 2009. Although the ISM fluctuates
during 2006 and the end of 2007, it does move above 50, indicating the majority
practitioners hold a positive opinion about the business condition during that stage.
Figure 3 indicates that the US crisis starts around the fall of 2008, when ISM is under
50 and sharply decreases.
Table 6 shows the estimation results of equations (6) and (7). We use ISM as a
regime transition indicator for the US crisis in the logistic transition function,
equation (7), to obtain the fitted value of G. LSTR1 model indicates that two regimes
exist in our sample period. The significant shape parameter 𝛾 is 4.111, implying a
switch from regime 1 to regime 2. 𝛽 represents the news effect of filtered volatility
during recession, and 𝛽 + 𝛽 ′ represents the new effect of filtered volatility during
expansion. From Table 6, the significant coefficient of consolidated news 𝛽 ′ also
indicates that the news effect is significantly different between the regimes 1 and 2.
Figure 4 plots fitted G and NBER dates in our sample range. The NBER date
shows the recession started in the fall of 2007 and ended in the beginning of 2009.
NBER only provides the information about the start and end dates, but ISM is
37
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
continuous which can be used in any sample to identify the business cycles.
According to the plot of fitted G in figure 4, the economy starts going through a
sluggish time from the beginning of 2007.
After a slight resurgence period in 2008,
the economy starts the next round recession in the fall of 2008 and end in the mid of
2009. Then the economy starts to recover in the fall of 2009. The STR provides a
more detailed timeline of the business cycles.
5.3 News Surprise Effects over Business Cycles
Table 7 presents estimation results of the news surprise effect on LOB
characteristics. Table 12 accumulates the number of significant news surprises in each
country. For volatility, 17% of 89 news categories are significant in the recession and
21% of 89 news categories are significant in the expansion. So volatility is more
affected by news surprise during the expansion. Besides US and EC news, volatility is
affected by news from German, Italy and Spain. With respect to other significant
news in the recession, housing related news announcements, such as Existing Home
Sales and NAHB Housing Market Index, positively affect volatility. This result is
supported by the fact that the US crisis originated from housing subprime crisis. News
announcements related to production, price index and employment, such as GDP
Annualized QoQ Advance, Core PCE QoQ and ADP Employment Change,
negatively affect volatility during the expansion. We find that some news
announcements have significantly different effects in both regimes, such as, Labor
Costs, NAHB Housing Market Index, ISM-Non-Manf. Composite, Nonfarm
38
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Productivity – Final and PCE Core MoM. This finding agrees with the
state-dependent effects documented by Ben Omrane & Savaser (2013), which shows
that the news effects vary with the economic states.
In Table 12, almost half of types of news are significantly related to depth as 45%
types of news in recession and 57% types of news in expansion are significant. So
depth is more affected by news surprise during an expansion. Besides the US and
Euro zone, news from Germany and Italy also has significant impacts on depth. News
announcements related to forward looking and monetary policy positively affect
depth during expansion, such as, Business Climate Indicator and FOMC Rate
Decision. We find that 24% of 89 news have significantly different effects in both
recession and expansion, indicating the news effects on depth are state-dependent.
Normally, depth will decrease around news announcements as conservative traders
will provide more limit orders with a “thin” book (Erenburg and Lasser, 2009).
However, our estimation shows that the sign of the news depends on the specific
content of the news. News related to housing markets are significant in both regimes:
New Home Sales, Pending Home Sale MoM, Housing Starts, Existing Home Sales
and
NAHB
Housing
Market
Index.
Personal-consumption-related
news
announcements cause depth fluctuations: Personal Consumption- Preliminary,
Personal Spending, and Private Consumption QoQ. Depth is significantly affected by
ISM Manufacturing and Consumer Confidence Index in both regimes, implying that
traders rely on the forward looking news to make decisions. Notably, some
39
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
state-dependent news announcements have opposite effects in two business regimes.
For example, the unemployment rate is negatively related to depth in a recession
while positively related in an expansion.
From Table 12, quoted spread is more affected by news surprise during
expansion.
Significant news are from EC US and GE. In the recession, news
announcements used as price index are significant, such as CPI Estimate YoY.
Income related news announcements positively affect quoted spread. In the expansion,
forward looking news has negative effects on quoted spread, such as consumer
confidence index. The quoted spread negatively reacts to news related to personal
consumption: Personal Consumption and Personal Spending. Housing market news
also affects quoted spreads: Housing Starts, Pending Home Sales MoM. For quoted
spread, 13% of news announcements have state-dependent effects as unemployment
rates is negatively related to quoted spread, regardless of regimes. Some of
state-dependent news announcements have opposite effects in two business regimes.
That is, the sign of coefficients of ISM Manufacturing, Housing Starts and Initial
Jobless Claims changes over different economic stages.
For slope, 21% of news announcements are significant in the recession while
only 3% of news is significant in the expansion, implying an asymmetric slope news
response to economic cycles.
During the recession, slope is mainly affected by news
of Euro Zone, Germany and US while in the expansion, only US news affects slope.
News announcements related to forward looking, employment, monetary policy are
40
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
significant during the recession, such as ISM Manufacturing, FOMC Rate Decision,
and Unemployment Rate. In the expansion, news related to income has negative
impacts on slope, such as Change in Nonfarm Payrolls.
Table 7 also exhibits the estimation results of unscheduled news related to crisis
in the case of news surprise. US Unscheduled news has oppositely significant effect
on all characteristics. EC unscheduled news also has significantly different effect on
all characteristics.
5.4 Pure News Effects over Business Cycles
Table 8 shows the estimation results of pure news effects. Table 12 panel B
shows the percentage of significant pure news for the characteristics. The total
number of significant news in equation (9) is more than that in equation (8) although
only 74 news announcements are estimated in equation (9). This is caused by the
difference in regression objectives between pure news, and its news surprise. While
the surprise regression captures the magnitude of news effect, while the pure news
regression tries to capture the number of significant news announcements. The 13%
difference occurs because there is no surprise when actual equals expectations.
Contrary to surprise, more types of news are significant during expansion in the
case of pure news. During the expansion, significant news are primarily in the US. In
euro countries, volatility is more affected by German macro news during the recession.
News positively affects volatility during the recession, such as, ECB announcement of
interest rates, the University of Michigan Confidence preliminary and consumer
41
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
confidence index. Volatility also responds to news related to the business
environment during the expansion: Philadelphia FED business outlook and IFO
business climate. We find that 10% of news announcements have state-dependent
effects. The release of FOMC rate decisions triggers a growth in volatility in both
regimes. Similar to news surprise, housing starts and existing home sales all
positively affect volatility in both regimes.
Depth in equations (8) and (9) both react to more than half of the news categories.
Similar to the case in (8), more types of pure news affect depth during expansion
where 71% of news are significant. Similar to the case in (8), significant news are
evenly distributed among all the regions in expansion period. Depth responds to 47%
of pure news with state-dependent effects.
In general, news announcements related
to housing markets, monetary policy, price index, personal consumption, income, and
forward looking have significantly positive effects on depth in both regimes. For
instance, the occurrence of housing starts, new home sales and pending home sales
MoM leads to a significant increase in depth. And depth experiences a growth in
recession and expansion when initial jobless claims, IFO business climate, PCE core
MoM are released.
According to Table 12, for the quoted spread in equation (9), the pure news effect
in recession (30%) is weaken than in expansion (42%). Similar to the other
characteristics, the quoted spread is strongly affected by news that are related to
housing market: new home sales, housing starts and pending home sales MoM have
42
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
significant negative effects on quoted spreads. The quoted spread decreases in
response to the release of price index during the recession. We find that 17% of news
announcements have state-dependent effects and almost all of them are negatively
related to quoted spread. News related to credit, employment and customer
confidence have significantly different effects in recession and expansion. Notably,
ISM Manufacturing has a significant opposite effect in recession versus expansion.
Slope is rarely significant for any news category but it has larger response to pure
news during recession compared to expansion. In general, the pure news effect on
slope is weaker than the news surprise. All significant news are positively related to
the slope during recession except the business climate indicator. The release of New
Home Sales is positively related to slope during expansion. Note that the sign of
coefficient of Chicago Purchasing Manager changes over business cycles while
Average Hourly Earnings MoM positively affects slope in both regimes.
In summary, two types of news are likely to affect four characteristics in both
regimes. The first type is strongly related to the crisis context. For instance, among
the four characteristics, the frequency of significant news related to housing market is
higher than that of the other type of news, which may be related to the fact that the
crisis originated from the US subprime mortgage market. The other type is the news
that are economic indicators, such as ISM manufacturing, unemployment rate and
personal consumption.
43
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 8 also exhibits the estimation results of unscheduled news related to the
crisis in the case of pure news. Both US and EC unscheduled news trigger a decline in
depth and slope during the recession but has a positively effect during the expansion.
This asymmetrical effect was also observed in Table 7.
5.5 Asymmetric News Effects over Business Cycles
Table 9 presents the estimation results of equation (10). Aggregate good news has
stronger effect on depth. Also, aggregated Good news have state-dependent effects on
volatility, depth and quoted spread. However, in the case of slope, we have significant
aggregated good news during recession and significant aggregated bad news during
expansion. When aggregated good news or bad news happens, volatility, slope and
depth increases, but quoted spread decreases. Compared coefficients of the two
regimes, the characteristics tend to have a more intense response to the news during
expansion.
US unscheduled news has an asymmetrical effect on volatility and depth during
both regimes. US unscheduled news also has an asymmetrical effects on depth and
spread, indicating that spread increases (decreases) during recession (expansion)
respectively. EC unscheduled news is also asymmetric on slope depth and quoted
spread. Depth tends to decrease during the recession but increase in the expansion.
Slope tends to decrease during the recession and increase during the expansion when
unscheduled news occurs.
44
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
6. Robustness Check
In this section, we perform robustness tests for the results presented in the above
section. First, to check whether the proxy choice of method for different
characteristics affects the empirical results, we summarize the measures of
characteristics used in the literature. Basically, the proxies and their calculation
methods for the LOB characteristics vary with the information extracted from the
LOB. Second, we investigate the news effect on the ask side and bid side of the LOB.
We utilize the methods of depth and slope in section 3 to investigate the effect of
news surprise on the ask side and bid side of the LOB. Third, we examine the news
effects on the different levels of the LOB. We perform the effects of news surprise on
the volatility at the 2nd to 5th levels and the 5th to 10th levels in the book. And we
perform the effects of news surprise on the depth and slope at the 2nd to 5th levels and
the 5th to 10th levels on the ask side and bid side of the book.
In this section, we show alternative measures of characteristics for tick-by-tick
data in section 6.1. Then we show the estimation results of the robustness check on
the alternative methods for slope and volatility in section 6.2. Next, we perform the
news effects on the LOB at the different sides (ask side and bid side) in section 6.3.
Finally we show the news effects on the LOB at the different levels (the 2nd to 5th
levels and the 5th to 10th levels) in section 6.4.
45
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
6.1 Alternative Measures of Characteristics
Alternative Depth Method
Instead of the depth measure we introduced in section 3.1.1, an alternative is
widely applied in the literature which uses the term “depth” as the quantity of
liquidity offered and demanded in the LOB (Nigmatullin, Tyurin and Yin, 2007;
Gunther, W. 2008, Biais et al. 1995; Cao et al., 2004).
The LOB depth is the size which corresponds to the price of each level in every
tick in LOB. In the literature, the depth for an interval n in the LOB is the sum of all
the sizes for each tick at both sides. We compile the market depth based on the total
number of limit orders posted at the bid and ask prices at the end of each time interval.
𝐴
So the size in every tick of an interval is calculated as: 𝑆𝑖𝑧𝑒𝑛,𝑖 = ∑𝐿𝑙=1[𝑄𝑛,𝑖,𝑙
]+
𝐵
∑𝐿𝑙=1[𝑄𝑛,𝑖,𝑙
]. So, the alternative method of “depth” is the overall size in a tick
measures the amount of LOB liquidity. Compared to the depth used in the section 3,
this alternative does not contain the prices. Following the same method introduced in
section 3, we denote 𝑆𝑖𝑧𝑒𝑡,𝑛 as the time-weighted average of size in interval n at day
t. Following the filter procedures in section 3.5, IAOM is applied to obtain the
dummy which controls the seasonality pattern in a regression. Denote the seasonality
𝑆𝑖𝑧𝑒
dummy for size-weighted spread 𝑆𝑖𝑧𝑒𝑡,𝑛 as 𝐴𝑉𝑡,𝑛
for interval n at day t. The
intraday pattern of size is shown Figure 5.e, which has a similar tendency of intraday
pattern to depth in Figure 1.a.
46
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Alternative Spread Method
Usually, the spread of LOB refers to the price difference between the best quotes.
For example, the quoted spread in section 3 is the percentage of the difference
between the best quotes relative to the mid-price. Nevertheless, Kozhan and Salmon
(2010) use size-weighted average prices at the ask side and bid side to get the
size-weighted spread. Compared to the measures illustrated in section 3,
Size-weighted spread combines all the prices and sizes for all LOB levels. Followed
by the size-weighted prices of the ask and bid side in section 3.1.4, we denote the
size-weighted spread in one tick i in an interval n as 𝑤𝑠𝑝𝑟𝑒𝑎𝑑𝑛,𝑖 : 𝑤𝑠𝑝𝑟𝑒𝑎𝑑𝑛,𝑖 =
𝐴
𝐵
𝐴𝑃𝑛,𝑖
− 𝐴𝑃𝑛,𝑖
. Following the same method introduced in section 3, we denote
𝑤𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 as the time-weighted average of size-weighted spread in interval n at day
t. Following the filter procedures in section 3.5, we use IAOM to form a controlling
dummy of the seasonality in regression, and denote the seasonality dummy for
𝑤𝑠𝑝𝑟𝑒𝑎𝑑
size-weighted spread 𝑤𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 as 𝐴𝑉𝑡,𝑛
for interval n at day t. The intraday
pattern of size-weighted spread is shown Figure 5.c, which has a similar intraday
pattern as the quoted spread in Figure 1.b.
Alternative Slope Method
Instead of the slope measure in Section 3, two alternative measures of slope are
summarized from literature to investigate robustness of the results. Similar to the
other characteristics, these slope measures differ based on the amount of LOB
information. The three measures were implemented in two papers. “NORM SLOPE”,
47
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
the slope measure introduced by the Naes & Skjeltorp (2006), and “WSLOPE”, the
slope measure by Kozhan &Salmon (2010).
a.
“NORM SLOPE”
With respect to the second alternative measure, we use “NORM SLOPE” by Naes
& Skjeltorp (2006) where the slope of LOB is normalized to the total size at that tick.
The difference of “NORM SLOPE” and the measure in Section 3.1.3 is that “NORM
SLOPE” is the elasticity of size to price for the book. But the slope of the best quotes
in Section 3 is measured based on the ratio of size to the percentage change of price.
In other words, “NORM SLOPE” is the percentage size at each price level in one tick
relative to the total size of all the price levels in that tick. Hence, the first and second
terms here are measured in the same units in this method. In equation (12), NORM
SLOPE standardizes the order book to the market cap and the corresponding liquidity
in the LOB. Differing from the slope in Section 3, it can be used for comparisons
among LOB levels (for example, the comparison between the slope of the best and the
rest of the quotes). Define the percentage size of total size at each price level l in
𝑄𝐴
𝐴
one tick i at ask side as 𝑅𝑄𝑛,𝑖,𝑙
, so this percentage is calculated as: 𝑅𝑄 𝐴𝑛,𝑖,𝑙 = ∑𝐿 𝑛,𝑖,𝑙
.
𝑄𝐴
1
𝑛,𝑖,𝑙
The NORM SLOPE for the ask side on each tick 𝑖 in interval n on date t is
1
𝑅𝑄𝐴
𝑛,𝑖,1
𝑁𝑂𝑅𝑀
𝐴𝑠𝑘𝑠𝑙𝑜𝑝𝑒𝑛,𝑖
= 𝐿 [ 𝑃𝐴
𝑛,𝑖,1
−1
𝑚𝑖𝑑𝑛,𝑖
+ ∑𝐿−1
𝑙=1 |
𝑅𝑄𝐴
𝑛,𝑖,𝑙+1
−1
𝑅𝑄𝐴
𝑛,𝑖,𝑙
𝑃𝐴
𝑛,𝑖,𝑙+1
−1
𝑃𝐴
𝑛,𝑖,𝑙
|],
(12)
similarly we get the NORM SLOPE for the ask side on each tick i in interval n on
𝑁𝑂𝑅𝑀
date t :𝐵𝑖𝑑𝑠𝑙𝑜𝑝𝑒𝑛,𝑖
. The reason for taking the absolute value of the NORM SLOPE
48
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
for best quote is that we want to capture the magnitude of the elasticity. Finally, the
𝑁𝑂𝑅𝑀
“NORM SLOPE” for tick 𝑖 in interval n at day t is: 𝑠𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
=
1
𝑁𝑂𝑅𝑀
𝑁𝑂𝑅𝑀
(𝐴𝑠𝑘𝑠𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
+ 𝐵𝑖𝑑𝑠𝑙𝑜𝑝𝑒𝑡,𝑛,𝑖
) × 2.
So “NORM SLOPE” for a tick is the
average of the NORM SLOPE of ask and bid side at that tick. Using the same method
𝑁𝑂𝑅𝑀
in section 3, the NORM SLOPE at interval 𝑛 in day 𝑡 is 𝑠𝑙𝑜𝑝𝑒𝑡,𝑛
. We obtain the
𝑛𝑜𝑟𝑚𝑠𝑙𝑜𝑝𝑒
seasonality dummy of NORM SLOPE as 𝐴𝑉𝑡,𝑛
for interval n at day t. The
intraday pattern of NORM SLOPE is shown Figure 5.a, which has a similar intraday
pattern to in Figure 1.c.
b. WSLOPE
The slope measure used by Kozhan and Salmon (2010) only considers the best
quotes and the second best quotes. We label this slope as “WSLOPE”. The slope can
be interpreted as the percentage of the difference of the best quote price to the second
quote price relative to the best quote size.
In other words, the difference of prices is
normalized by the size of the best quote. Then we obtain the slope for the demand
and supply curve. The WSLOPE of the ask side of the LOB at tick 𝑖 is denoted as
𝐴 12
𝑒𝑛,𝑖
:
𝐴
𝑒𝑛,𝑖
=
𝐴
𝑃𝐴
𝑛,𝑖,1 −𝑃𝑛,𝑖,2
𝑄𝐴
𝑛,𝑖,1
.
(13)
𝐵
Similarly the WSLOPE of the bid side of the LOB at tick 𝑖 is denoted as 𝑒𝑛,𝑖
.So
1
𝐴
𝐵
the slope at tick 𝑖 in interval n on day t is: 𝑒𝑡,𝑛,𝑖 = (𝑒𝑡,𝑛,𝑖
+ 𝑒𝑡,𝑛,𝑖
) × . The slope by
2
Kozhan and Salmon (2010) is the percentage of the change of prices between the best
12For
the readability of the results, the price used in calculation of slope by Kozhan and Salmon is in the basis
point. And the size is in the million.
49
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
and the second best relative to the size at the best quote. As an alternative slope
measure, this slope is applied in the estimation in order to compare with other slopes
that combine more information. We then follow the same method as in section 3
where the slope at interval 𝑛 in day 𝑡 is following: 𝑒𝑡,𝑛 where 𝑒𝑡,𝑛 is the
time-weighted average of the slope in interval n in day 𝑡. Following the filter
procedures in section 3.5, we use IAOM to form a dummy for controlling seasonality
in the regression, and denote the seasonality dummy for slope by Kozhan and Salmon
𝑒
(2010) 𝑒𝑡,𝑛 as 𝐴𝑉𝑡,𝑛
for interval n at day t.
Alternative Method of Return
The alternative measure of return is based on an alternative price. The alternative
price is the average of the best ask price and the best bid price in every tick in the
1
𝐴
𝐵
𝐴
interval: 𝑚𝑛 = 2 (𝑃𝑛,𝛾
+ 𝑃𝑛,𝛾
), where 𝑃𝑛,𝛾
is the best ask price of the last tick 𝛾𝑛
𝑛
𝑛
𝑛
𝐵
in interval n and 𝑃𝑛,𝛾
is the best bid price of the last tick 𝛾𝑛 in interval n. Then
𝑛
calculate the difference of the log-price of the last observation between the last and
current interval to obtain the return.
So the alternative method of return
𝑟𝑡,𝑛 is: 𝑟𝑡,𝑛 = (log(𝑚𝑡,𝑛+1 ) − log(𝑚𝑡,𝑛 )) × 10013 . Here the volatility is a 5-min
centered return, that is,|𝑟𝑡,𝑛 − 𝑟̅ | where 𝑟𝑡,𝑛 is the return for interval n in day t, and
𝑟̅ is the average return for 𝑟𝑡,𝑛 of whole sample. Then we filter the volatility
|𝑟𝑡,𝑛 − 𝑟̅ |, denoted here as Abs_ret, by FFF following the procedures in section 3.
Finally, we get the filtered volatility based on the best quote: 𝑦𝑡,𝑛 .
13
Noting that the return calculated this way is very small, it is multiplied by 100 to give its percentage value.
50
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
6.2 Robustness Check of Characteristics
To describe the condition of LOB, we select four categories of LOB
characteristics, which are “slope”, “spread”, “depth” and “volatility”. Many
alternative methods are introduced to compute the four categories of characteristics.
We chose one method from each category to construct the VAR-STR model in the
case of a news surprise.
In Appendix A, an example is given to illustrate that the
VAR model can be constructed based on all the alternative combinations of different
characteristics.
To show the robustness of the slope, we compare the VAR-STR results of slope
𝑁𝑂𝑅𝑀
( 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 ), NORM SLOPE ( 𝑠𝑙𝑜𝑝𝑒𝑡,𝑛
) and WSLOPE ( 𝑒𝑡,𝑛 ) with other
characteristics unchanged ( 𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 , 𝑄𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 , 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 ). Table 10 show the
estimation results of the VAR-STR model for news surprise with three slope
measures14.
Slope is calculated the same as NORM SLOPE, except that NORM SLOPE is
normalized by size. Slope and NORM SLOPE have similar total number of
significant news in recession but NORM SLOPE has a larger response to news during
expansion. The significant news in the estimation of slope and NORM SLOPE are
distributed among EC, GE, IT and US. The news that have asymmetric
state-dependent effects in the case of NORM SLOPE is much larger than that for
slope. The reason may related to the calculation of NORM SLOPE which avoids the
14
We eliminate the estimation results of the other three characteristics.
51
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
“different unit” problem in its calculation. News related to personal consumption and
business indicator: preliminary and ISM manufacturing are both significant in the
case of slope and NORMSLOPE
WSLOPE is calculated by using only the best and the second best quotes in the
LOB, while slope is constructed based on all the information in the LOB. In Table 13,
the significant estimation results of WSLOPE is inferior to slope as only 15% news
are significant in WSLOPE while 25% are significant in slope. The results indicate
that the news effect does exist no matter which method is chosen and that the slope
which is calculated based on the whole book data is more informative. This is agrees
with the empirical results in Cao et al. (2004), who shows that the LOB is more
informative than the LOB’s best quote.
To show the robustness of our volatility measures, we estimates the two methods
of volatility (𝑣𝑜𝑎𝑙𝑡𝑡,𝑛 and 𝑦𝑡,𝑛 ) with (𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 , 𝑄𝑠𝑝𝑟𝑒𝑎𝑑𝑡,𝑛 , 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛 ) in the model
VAR-STR with a news surprise. Table 11 presents the estimation results of the
original volatility (section 3) and the alternative volatility introduced in section 6. The
difference between methods is that the original volatility exploits all the information
in LOB, while the alternative method use only the best quote. In Table 13, 𝑦𝑡,𝑛 is
only significant for 20% of news while the original volatility is 40%. Although the
original volatility has a similar response during recession (24%) or expansion (21%),
the alternative volatility 𝑦𝑡,𝑛 has much stronger response to expansion (16%),
relative to recession (6%).
News related to housing market and employment are
52
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
both significant for both volatilities, such as housing starts, NAHB housing market
index, and initial jobless claims. The results indicate that the news effect does exist no
matter which method is been chosen; however, the original volatility is more
informative than the alternative, implying the levels beyond the best quote are
informative.
Overall, the robustness results for different measures of volatility and slope
indicate that news effects on the LOB over regimes are robust to different measures of
these proxies.
In addition, the news effect is stronger when measures are constructed
based on the whole book.
6.3 Robustness Check for News Effect on Ask and Bid Sides in LOB
In this section, we show the news effect on depth and slope for ask side and bid
side in the LOB. The methods of depth and slope are introduced in section 3. Table 14
presents the news surprise effect on depth at the ask side and bid side in the LOB. In
terms of the number of significant news announcements, news surprise has stronger
effect on depth at both sides during the expansion. For example, in the ask side, 50
out of 89 news announcements are significant during the expansion but only 37 news
announcements are significant during the recession. Also, nearly one-third news
announcements have state-dependent effect or sign-switch effect at the ask side and
the bid side. Table 15 presents the news surprise effect on slope at the ask side and the
bid side of the LOB. In terms of the number of significant news announcements, news
surprise has stronger effect on depth at the bid side. 15 out of 89 news announcements
53
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
are significant on the bid side but only 9 out of 89 news announcements are
significant on the ask side. No news announcement has the state-dependent effect or
sign-switch effect on slope for both sides. This result is the same as the estimation
result of slope in Table 7.
In terms of the estimation coefficient, news surprise has similar effects on depth
at the ask side and the bid side. For instance, in Table 14 Panel A, Household Cons
QoQ – Preliminary has negative effect on depth during the recession, but it has
positive effect on depth during the expansion for both sides in the LOB. And some
news announcements, which have significant effect on depth of the whole book
(Table 7), also have significant effect on depth at the ask side and the bid side, such as
Consumer Confidence Index, Construction Spending MoM and Personal Spending.
This result indicates that news surprise effect is robust to the different sides of the
LOB.
6.4 Robustness Check for News Effect on different levels in LOB
To investigate the news effect on the different levels in the LOB, we estimate the
news effect on volatility, depth and slope at the 2nd to 5th levels and the 5th to 10th
levels in the book. We utilize the methods of volatility, depth and slope that we
introduced in section 3. Table 16 shows the news surprise effect on volatility at the
2nd to 5th levels, the 5th to 10th levels of the book. News effect on the 2nd to 5th
levels of the LOB is slightly inferior to that at the 5th to 10th levels: 32 out of 89
news announcements are significant at the 5th to 10th levels, and only 29 significant
54
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
news announcements at the 2nd to 5th levels. In terms of the estimation coefficients,
new surprise has similar effects on volatility in the book. For example, Housing
Starts has negative effect during the recession at both the 2nd to 5th levels and the
5th to 10th levels of the LOB.
Table 17 and Table 18 show the news surprise effect on depth or slope at the 2nd
to 5th levels and the 5th to 10th levels at the ask side and the bid side of the book
respectively. The number of significant news announcements at the 5th to 10th
levels of the book is more than that at the 2nd to 5th levels, which indicates that the
upper levels of the LOB are more sensitive to news surprise. In terms of the
estimation coefficients, generally, the news effect is stronger on depth and slope at
the upper levels in the book. For example, New Home Sales has larger negative
effect on the 5th to 10th level for both recession and expansion. This result also
verifies that the upper levels of the LOB have important information. Also, some
news announcements, which have significant effect on depth for the whole book,
also have significant effect on depth for the different levels, such as Existing Home
Sales, NAHB Housing Market Index, and PPI Ex Food and Energy MoM. In sum,
news surprise effect is robust to the different levels of the LOB.
55
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
7. Conclusion
We investigate the dynamics of LOB characteristics in FX ECN markets with
respect to macroeconomic news between Jan. 3rd 2006 and Dec. 31st 2009, during
which we the US crisis starts in 2008. We apply a VAR-STR model to
macroeconomic news over different business regimes and find that the effect on
characteristics not only vary with the type of news but also vary with the different
business regimes.
We find that four characteristics are significantly influenced by news
announcements but they can respond to economic cycles asymmetrically. Slope is
more affected by both the news surprise and pure news during the recession; depth is
more affected by pure news and the news surprise during the expansion. Quoted
spread is more affected by the news surprise during expansion while has more intense
responses to pure news to recession. News surprise has stronger effect on volatility
during recession while pure news strongly affects volatility during expansion. Pure
news announcements have stronger effects on characteristics, compared to news
surprise. In addition, the LOB characteristics tend to have a more intense response to
aggregated good or bad news during the expansion.
News announcements related to monetary policy, personal consumption, price
index, forward looking, and employment significantly affect the four characteristics
over different economic stages. Furthermore, news related to housing market,
economic indicator consistently affects B characteristics. In addition, some news
56
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
announcements exhibit state-dependent effects as some of them have opposite LOB
characteristics effects in two business regimes. Therefore, we find that news effect on
LOB characteristics is affected by the context of the recent global crisis.
Our results show that news effects on LOB characteristics in different regimes is
partially robust among different alternatives measures of these characteristics. The
robustness check on depth and slope at ask side and bid side indicate that news
announcements affect both sides of LOB symmetrically. Moreover, from the
robustness check on different levels in the LOB, we find that upper levels in the LOB
are more sensitive to news announcements that lower levels in the LOB.
57
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Reference
Ahn, H.-J., K.-H. Bae, K. Chan, (2001). “Limit Orders, Depth, and Volatility:
Evidence from the Stock Exchange of Hong Kong”. Journal of Finance, 56,
767-788.
Andersen, T. G., and Bollerslev, T. (1998). “Deutsche Mark-Dollar Volatility: Intraday
Activity Patterns, Macroeconomic Announcements, and Longer Run
Dependencies.” The Journal of Finance, 53. 219-265.
Andersen, T. G. (2003). “Micro effects of macro announcements: Real-time price
discovery in foreign exchange”, The American Economic Review, 93, 39-63
Andersen, T.G., Bollerslev, T., Diebold, F.X. and Vega, C. (2007): “Real-Time Price
Discovery in Stock, Bond and Foreign Exchange Markets”. Journal of
international Economics, 73, 251-329
Balduzzi, P., Elton, E.J. and Green, T.C. (2001), “Economic news and bond prices:
Evidence from the US Treasury market”, Journal of Financial and Quantitative
Analysis, 49, 523–543.
Bauwens, L., Giot. P, (2001), Economic Modeling of stock market intraday Activity,
(Klwwer Academic Publishers).
Ben Omrane, W. and Savaser, T.
(2013), “The Sign Switch Effect of
Macroeconomic News in Foreign Exchange Markets”, MFS 2013 Working
Paper.
Bauwens, L., Omrane, W., Giot P., (2005). "News announcements, market activity
and volatility in the euro/dollar foreign exchange market," Journal of
International Money and Finance, 24, 1108-1125.
Beltran Lopez, H., Durre, A. Giot, P., (2004) “How does liquidity react to stress
periods in a limit order market?” National Bank of Belgium 2004 working
paper, 49.
Beltran Lopez, H., Helena, Menkveld, Albert J., Grammig, J., (2012). “Limit order
books and trade informativeness”, The European Journal of Finance, 18,
737-759.
58
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Ben Omrane, W., Bodt. E, (2007). "Using self-organizing maps to adjust for intra-day
seasonality," Journal of Banking & Finance, 31, 1817-1838.
Bessembinder, H., (2002). “Trade Execution Costs and Market Quality after
Decimalization” The Journal of Financial and Quantitative Analysis, 38, 4,
747-777.
Biais, B., Hillion, P., and Spatt, C. (1995). “An Empirical Analysis of the LOB and
the Order Flow in the Paris Bourse”. Journal of Finance, 50, 1655-1689.
Cao, C, Hansch, O, and Wang, X. (2004), “The informational content of an open
LOB”. EFA 2004 working paper.
Coppejans, M., Domowitz, I., and Madhavan, A., (2004), “Resiliency in an
Automated Auction”, Unpublished Working Paper.
Duong, H.N., and Kalev, P.S., (2009). “Order Book Slope and Price Volatility”,
Unpublished Working Paper.
Erenburg, G., and Lasser, D., (2009). “Electronic limit order book and order
submission choice around macroeconomic news”, Review of Financial
Economics, 18, 172-189
Evans, M. D., and Lyons, R. K. (2008). “How is macro news transmitted to exchange
rates?” Journal of Financial Economics, 88, 26-50
Fratzscher, M. (2009). “What explains global exchange rate movements during the
financial crisis?” Journal of International Money and Finance, 28, 1390-1407
Goldstein, Michael A. and Kavajecz, Kenneth A., (2004). “Trading strategies during
circuit breakers and extreme market movements,” Journal of Financial Markets,
7, 301-333
Gunther, W. (2008). “The impact of liquidity shocks through the limit order book”,
Venter of Financial Studies (CFS) working paper No.2008/53
Kim, J.W., Lee, J., Morck, R., (2004). “Heterogeneous Investors and their Changing
Demand and Supply Schedules for Individual Common Stocks.” NBER
Working Paper Series No.10410
Kozhan, R., & Salmon, M. (2010). “The information content of a LOB: The case of
an FX market.” Journal of Financial Markets, 15, 1-28
59
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Laakkonen H, Lanne M. (2010) “Asymmetric News Effects on Exchange Rate
Volatility: Good vs. Bad News in Good vs. Bad Times.” Studies in Nonlinear
Dynamics & Econometrics, 14, 1-36.
Love R. and Payne R. (2008) “Macroeconomic News, Order Flows, and Exchange
Rates”. Journal of Financial and Quantitative Analysis, 43, 467-488
Mancini L, Ranaldo A, and Wrampelmeyer J. (2012) “Liquidity in the Foreign
Exchange Market: Measure, Commonality, and Risk Premiums” Journal of
Finance, 67, 1805-1841.
Melvin, M. and Taylor, M.P. (2009) “The crisis in the foreign exchange market”. 28,
CESifo Working Paper Series No. 2707
Naes R. and Skjeltorp, J. A. (2006). “Order book characteristics and the
volume-volatility relation: Empirical evidence from a limit order market”.
Journal of Financial Markets, 9, 408-432.
Nigmatullin E., Tyurin K. and Yin H. (2007). “Heterogeneous VAR Dynamics of
Limit Order Book Depth, Trade Imbalance, and Volatility on the NYSE”. The
NBER’s Program of Research on Economic Fluctuation and Growth, 2007.
Riordan, R., Storkenmaier, A., Wagener, M., and Sarah Zhang, S. (2013). “Public
information arrival: Price discovery and liquidity in electronic limit order
markets.” Journal of Banking & Finance, 37, 1148-1159.
Robert F. Engle, Fleming M., Ghysels E., and Nguyen G. (2012) “Liquidity, Volatility
and Flights to Safety in the U.S. Treasury Market: Evidence from a New Class
of Dynamics Order Book Models”. Journal of Economic Literature,
590.138-178.
Russell, Jeffrey R. and Engle, Robert F., (1998) “Autoregressive Conditional Duration:
A New Model for Irregularly Spaced Transaction Data”. The Econometric
Society 66 1127-1162.
Teräsvirta, T. (1994). “Specification, Estimation, and Evaluation of Smooth Transition
Autoregressive Models”. Journal of the American Statistical Association, 425.
208-218.
Veredas, D., (2006) “Macroeconomic surprises and short-tern behavior in bond
futures” Empirical Economic 30, 843-866.
60
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Tables
Table 1. Descriptive Statistics of LOB
Mean
Median
Max
Min
Std. Dev.
Number of Tick in an Interval at Ask Side
340
290
1134
1
233.25
Number of Tick in an Interval at Bid Side
340
289
1134
1
233.32
Number of Level in a Tick at Ask Side
9
9
22
1
2.18
Number of Level in a Tick at Bid Side
9
9
22
1
2.3
Size in the Best level at Ask Side
13,096,144
12,000,000
174,000,000
100,000
8,807,530
Size in the Best level at Bid Side
12,459,348
10,900,000
135,200,000
100,000
9,231,479
Sum Size of the LOB at Ask Side
75,900,682
76,400,000
238,200,000
1,000,000
26,762,696
Sum Size of the LOB at Bid Side
75,532,845
75,200,000
333,000,000
200,000
28,801,328
Spread
0.00018
0.00015
0.0049
0.00005
0.00014
Number of Tick in an Interval at Ask Side
289
299
572
1
142.77
Number of Tick in an Interval at Bid Side
289
299
572
1
142.8
Number of Level in a Tick at Ask Side
23
23
45
1
5.18
Number of Level in a Tick at Bid Side
25
24
49
1
5.68
Size in the Best level at Ask Side
3,381,720
3,200,000
70,000,000
100,000
2,347,155
Size in the Best level at Bid Side
3,433,205
3,700,000
61,000,000
100,000
2,358,282
Sum Size of the LOB at Ask Side
66,115,081
67,100,000
132,750,000
500,000
13,815,915
Sum Size of the LOB at Bid Side
67,090,793
67,500,000
190,333,334
1,000,000
14,414,446
Spread
0.000157
0.00013
0.00363
0.00001
0.00009
Panel A: January 2006
Panel B: April 2009
Notes: Table 1 shows the descriptive statistics of the ask side and bid side of LOB in Jan. 2006 and Apr. 2009. Spread is the
difference of the best ask price and the best bid price. Sum Size of the LOB at Ask side or Bid side is the accumulated size in
all the levels in each tick.
61
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 2. Summary Statistics of Characteristics of LOB
Size
Depth
Abs_ret (%)
Abs_return (%)
QSpread
Sizespread
Slope
WSLOPE
NORMSLOPE
Mean
165,000,000
227,000,000
0.03
0.02
1.34
0.12
27,326.24
0.19
82,895.52
Median
157,000,000
214,000,000
0.02
0.01
1.11
0.08
23,614.00
0.13
52,259.76
Maximum
681,000,000
918,000,000
8.08
11.41
39.41
34.07
1,395,764
69.52
1,655,872
Minimum
700,000
1,908,834
0.00
0.00
0.07
0.00
1,883.37
0.00
102.94
Std. Dev.
80,918,569
116,000,000
0.04
0.06
1.75
0.16
19,956.18
0.47
84,406.19
Skewness
0.87
1.07
50.66
95.29
13.22
83.84
9.01
62.93
2.26
Kurtosis
4.84
5.28
9,855.75
13,145.47
216.25
14,386.11
291.98
6,588.99
11.24
N
245,780
245,780
245,780
245,780
245,780
245,780
245,780
245,780
245,780
AC(1)
0.973***
0.977***
0.24***
0.34***
0.949***
0.485***
0.762***
0.12***
0.88***
AC(2)
0.963***
0.968***
0.231***
0.119***
0.923***
0.418***
0.755***
0.085***
0.867***
Notes: Table 2 presents summary statistics of all the alternative methods of the depth, quoted spread, slope and volatility in 5-min frequency from 3rd Jan. 2006 to 31st Dec. 2009. Important to
note is that characteristics in the Table 2 are not yet adjusted for intraday patterns. Quoted Spread, measured in basis points, is denoted as QSpread, which is the percentage of the price
difference between the best bid price and the best ask price accounted for the mid-price of the LOB. Sizespread is difference of the size-weighted best ask price and the size-weighted best bid
price. Depth is the sum of the product of size and price in each interval. Size is the sum of size in each interval. Slope measures the elasticity of the LOB supply and demand curve. NORM
SLOPE is normalized slope. WSLOPE is the percentage of spread, measured in basis point, accounted for the size in the unit of million. Abs_ret is the absolute value of 5-min return.
Abs_return is the absolute value of 5-min return which is calculated with size-weighted price. Both Abs_ret and Abs_return are presented in percentage. N is the observation number in the
sample. AC (1) and AC (2) are the autocorrelation coefficients of characteristics with lag 1 and lag 2 respectively. The null is that no autocorrelation between characteristics and its lag orders.
*** denotes the probability of insignificant figure is at 1% level.
62
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 3. Summary Statistics of Characteristics in VAR-STR Model
S_Size
S_Depth
S_QSpread
S_Sizespread
Volat (%)
Y (%)
S_NORMSLOPE
S_Slopw
S_WSLOPE
Mean
2.056
1.965
0.765
0.722
-0.335
-0.206
0.988
1.368
0.407
Median
1.952
1.860
0.632
0.507
0.058
0.365
0.623
1.181
0.271
Maximum
8.418
7.888
22.493
41.454
11.372
11.735
19.618
60.481
147.422
Minimum
0.009
0.016
0.040
0.006
-22.288
-12.173
0.001
0.094
0.000
Std. Dev.
1.000
1.000
1.000
1.000
2.43
2.815
1.000
1.000
1.000
Skewness
0.867
1.067
13.090
83.166
-1.158
-1.571
2.259
7.959
63.702
Kurtosis
4.848
5.292
211.178
14072.3
5.782
6.149
11.249
213.972
6647.47
245,780
245,780
245,780
245,780
245,780
245,780
245,780
245,780
245,780
AC(1)
0.973
0.977
0.949
0.485
0.088
0.085
0.88
0.762
0.762
AC(2)
0.963
0.968
0.923
0.418
0.071
0.06
0.857
0.755
0.755
N
Notes: Table 3 presents summary statistics of the characteristics used in estimation. The sample period is between 3rd Jan. 2006 to 31st Dec. 2009. Volat is the filtered absolute value of
5-min return which is calculated with size-weighted price. Y is filtered absolute value of 5-min return. Both Volat and Y are presented in percentage. S_size is the standardized size by
dividing of 5-min size by Std. Dev. of 5-min size in the sample. S_depth is the standardized depth by dividing 5-min depth by Std. Dev. of 5-min depth in the sample. S_qspread is the
standardized quoted spread by dividing 5-min quoted spread by Std. Dev. of 5-min quoted spread in the sample. S_sizespread is the standardized sizespread by dividing 5-min
size-weighed spread by Std. Dev. of 5-min size-weighted spread in the sample. S_NORMSLOPE is the standardized NORM SLOPE by dividing 5-min NORMSLOPE by Std. Dev. of
5-min NORMSLOPE in the sample. S_WSLOPE is the standardized WSLOPE by dividing of 5-min WSLOPE by Std. Dev. of 5-min WSLOPE in the sample. N is the observation
number in the sample. AC (1) and AC (2) are the autocorrelation coefficients of characteristics with lag 1 and lag 2 respectively. The null is that no autocorrelation between
characteristics and their lag orders. *** denotes the probability of insignificant figure is at 1% level.
63
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 4. Correlations between Characteristics in VAR-STR Model
Y
Y
Volat
S_SIZE
S_DEPTH
S_NORMSLOPE
0.590
***
-0.008
***
-0.007
***
0.003
***
***
Volat
S_SIZE
S_DEPTH
S_NORMSLOPE
S_QSPREAD
S_SIZESPREAD
S_SLOPE
S_WSLOPE
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-0.020
***
-0.019
***
-0.002
0.988
***
0.551
***
0.608
***
-
0.015***
0.030***
-0.013***
-0.006***
-0.056***
0.067***
-
S_SLOPE
0.006***
0.008***
-0.214***
-0.203***
-0.126***
-0.190***
-0.192***
***
***
***
***
***
***
-0.092
-0.084
-0.101
***
S_SIZESPREAD
0.020
-0.096
***
-0.043
0.010
-0.093
***
-
S_QSPREAD
S_WSLOPE
-0.016
***
-
0.001
0.085
0.100
-0.058
***
-
Notes: Table 4 shows the correlation between the characteristics of liquidity and volatility variables used in the VAR-STR model. Volat is the filtered absolute value of 5-min
return which is calculated with size-weighted price. Y is filtered absolute value of 5-min return. S_size is the standardized size by dividing of 5-min size by Std. Dev. of 5-min
size in the sample. S_depth is the standardized depth by dividing 5-min depth by Std. Dev. of 5-min depth in the sample. S_qspread is the standardized quoted spread by dividing
5-min quoted spread by Std. Dev. of 5-min quoted spread in the sample. S_sizespread is the standardized sizespread by dividing 5-min size-weighed spread by Std. Dev. of
5-min size-weighted spread in the sample. S_NORMSLOPE is the standardized NORM SLOPE by dividing 5-min NORMSLOPE by Std. Dev. of 5-min NORMSLOPE in the
sample. S_WSLOPE is the standardized WSLOPE by dividing of 5-min WSLOPE by Std. Dev. of 5-min WSLOPE in the sample. The null is that the correlation between any
two characteristics are zero. *** denotes the probability of insignificant figure is at 1% level.
64
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 5. News Announcement Filter
Total
VAR-STR-Surprise
%
News
VAR-STR-Pure News
%
Freq.
%
News
%
Country
Freq.
News
Freq.
EC
547
16
380
69%
15
94%
380
69%
14
88%
FR
132
3
113
86%
3
100%
113
86%
3
100%
GE
461
13
317
68%
13
100%
317
68%
10
77%
IT
240
8
197
82%
8
100%
197
82%
7
88%
PO
43
2
37
86%
2
100%
37
86%
2
100%
SP
132
4
99
75%
4
100%
99
75%
2
100%
US
1897
43
1156
61%
43
100%
1156
61%
36
84%
Total
3452
89
2299
67%
88
100%
2299
67%
74
83%
Notes: Table 5 shows the percentage of surprise and pure news after second round of data filtering. Country provide the
countries’ name corresponding to the news categories: EC- Euro Country, FR-France, GE-German, IT-Italy, PO-Poland,
SP-Spain and US-United States. VAR-STR-Surprise shows the summary of filtered news for the STR model after the filter
process in section 4.3.1. VAR-STR-Pure news shows the summary of filtered news for the STR model after the filter process
in section 4.3.1. News and Freq. show the number of news categories and releases through entire sample period.
65
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 6. Estimation Results of STR Model
ISM (EURUSD)
𝜸
𝛽
𝛼
***
32.374
-0.17
(1.724)
(22.5)
(0.034)
4.111
𝛽′
201.598
***
(59.968)
𝛼′
LSTR Type
𝑐𝑘
-0.396
LSTR1
0.496
(0.22)
(0.209)
Notes: Table 6 presents the parameter estimations in equation (6) and (7) by using ISM as transition variable.
EUSUSD denotes for Euro/Dollar.𝛽 and 𝛽 ′ are the coefficients of consolidated news in equation (6). ISM
(Institute of Supply Management) is manufacturing index for US business cycles. The number in bracket are the
standard errors. * denotes the probability of the figures which is not statistically significant at 10% level. ***
denotes the probability of the figures which is not statistically significant at 1% level. LSTR Type is defined in
equation (6). K=1 for LSTR1 model.
66
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 7. Estimation Results of News Surprise
VOLAT
CN
Scheduled News
DEPTH
QSPREAD
SLOPE
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone Macro News
EC
Govt Expend QoQ - Preliminary
-3.965*
-
-
0.684***
-
-
-
-
-
-
-
-
EC
Gross Fix Cap QoQ - Preliminary
-
-
-
0.737***
0.276*
0.63
-
-
-
-
-
-
EC
Household Cons QoQ - Preliminary
-
-
-
-0.654***
0.249**
0.00
-
-
-
-
-
-
EC
Labour Costs YoY
-1.337*
4.308*
0.05
-
-0.535***
-
-
-
-
-
-
-
Retail Sales MoM
-1.488**
-
-0.092*
-0.185***
0.1
-
-
-
-
-
-
-
0.294***
-
0.233*
-
-
-
0.274***
-
0.21*
-
-
-
***
-
-
EC
EC
Trade Balance SA
EC
Business Climate Indicator
EC
CPI Core YoY – Final
EC
GDP SA QoQ – Final
EC
Gross Fix Cap QoQ – Final
EC
EC
EC
EC
Industrial New Orders SA (MoM)
Industrial Production SA MoM
ZEW Survey Expectations
CPI Estimate YoY
-
-
-
-0.178
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
0.276**
-
-
-
0.832***
-
-
-
-
0.207**
-
-0.116*
-
-
0.095**
-0.277***
-
-0.102**
0.00
-
-
-0.135*
-
-0.271
**
-
-
0.155
**
-
-
-0.443***
-
-
-
-
-
-
-
-0.205*
-
-
Notes: Table 7 presents the estimation of scheduled and unscheduled macro news effect that over different regimes of business in our sample period, 𝜃𝑞 in equation (8). CN presents the
corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. . Quoted Spread is denoted as QSpread, which is
the percentage of the price difference between the best bid and the best ask price accounted for the mid-price of the LOB. Volat is the filtered absolute value of 5-min return which is calculated
with size-weighted price. Depth is the sum of the product of size and price in each interval. Slope measures the elasticity of the LOB supply and demand curve.*, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and
recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession.
67
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 7. Estimation Results of News Surprise (continued)
VOLAT
CN
Scheduled News
DEPTH
QSPREAD
SLOPE
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Panel B: US Macro News
US
PPI Ex Food and Energy MoM
-
-1.605**
-
0.48***
-0.705***
0.00
-0.324***
-
-
-
-
-
US
ADP Employment Change
-
-3.459**
-
-
-1.691***
-
-0.619***
1.381***
0.00
-
-
-
US
PPI MoM
-
-
-
-0.2***
1.102***
0.00
-
-0.707***
-
-
-
-
US
Unemployment Rate
-
-
-
-0.143***
0.351***
0.00
-0.233***
-1.157***
0.00
0.194*
-
-
-
0.253***
-
-
-0.226*
-
-
-
0.151***
0.866***
0.00
-
-
-
-
0.942***
-
-
-0.097***
-
-
-
-
-
-
-
-
-
-
0.314***
-
-
-
-
0.151**
-
-
-
0.254***
-0.389***
0.00
-
-
-
-
-
-
-
0.412***
-0.92***
0.00
-
-
-
-
-
-
-
-0.105**
0.42***
0.00
0.446***
-0.828***
0.00
-
-
-
0.759***
0.00
-
-
-
-
-
US
US
US
US
US
US
US
US
Empire Manufacturing
0.908*
Existing Home Sales
1.277**
Factory Orders
1.112***
FOMC Rate Decision
GDP Annualized QoQ - Advance
GDP Annualized QoQ - Preliminary
Housing Starts
Initial Jobless Claims
-
-
-
-
-3.673***
-
-3.881**
-1.462***
-
-
-
-
-
-
-
-
-
-
-0.188**
1.58**
-6.649***
0.0
0.332***
-2.141***
0.00
-0.188**
-
-
0.3*
US
ISM Non-Manf. Composite
US
NAHB Housing Market Index
1.975***
-1.961***
0.0
-
-0.263***
-
-
-
-
-
-
-
US
Nonfarm Productivity - Final
1.39**
-3.725**
0.02
-
-0.622***
-
-
-
-
-
-
-
US
Nonfarm Productivity - Preliminary
1.542*
1.256*
0.61
0.463***
0.181***
0.54
-
-
-
-
-
-
Notes: Table 7 presents the estimation of scheduled and unscheduled macro news effect that over different regimes of business in our sample period, 𝜃𝑞 in equation (8). CN presents the
corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of
news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and recession. A significant
value of P_diff indicate that the coefficients are statistically different over expansion and recession.
68
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 7. Estimation Results of News Surprise (continued)
VOLAT
CN
US
US
US
Scheduled News
Avg Weekly Hours Production
Business Inventories
Chicago Purchasing Manager
Recession
-
DEPTH
Expansion
-
P-diff
Recession
-
-0.234***
-
-0.129***
QSPREAD
Expansion
SLOPE
P-diff
Recession
-
-
0.27***
Expansion
-
0.405***
0.00
-
-
-0.088*
0.716***
0.00
0.659***
P-diff
Recession
Expansion
P-diff
-
-
-0.771***
-
-
-
-
-
-
-
-
-
-
-
-
0.00
-
-
-
-
-
-
US
Construction Spending MoM
-
-
-
0.218***
US
Consumer Confidence Index
-0.688*
-
-
0.071**
0.517***
0.00
-
-1.488***
-
-
-
-
US
Core PCE QoQ - Advance
-
-8.989**
-
0.348***
-1.275***
0.00
-1.002***
1.7***
0.00
-
-
-
US
Core PCE QoQ - Preliminary
-
-
-
-
-2.393***
-
0.618***
1.452***
0.00
-0.631***
-
-
US
CPI Ex Food and Energy MoM
-
-
-
-0.071*
0.331***
0.00
-
-
-
-
-
-
-
0.478***
-
-
-
-
-
-
-
-
-
0.39***
-
0.147***
0.259*
0.25
-
-
-
-
-0.081*
0.426***
0.00
0.222***
-0.295***
0.00
0.273**
-
-
-
1.025***
-
-
-
-
-
-
-
-
0.305***
0.515***
0.00
-
-
-
0.00
-0.108*
-
-
-
-
-
-
-
-
-
US
US
US
US
US
US
US
US
Import Price Index MoM
Industrial Production MoM
ISM Manufacturing
ISM Milwaukee
Retail Sales Ex Auto MoM
Trade Balance
Personal Spending
Philadelphia Fed Business Outlook
0.858*
-
-
-
-
-
-
0.2***
-0.286***
-
0.182***
0.383***
0.00
-
-0.426***
-
-0.095**
0.389***
0.00
-
-
-
-
-
-
-
-
-
-
-
US
Avg Hourly Earning MOM Prod
-
-
-
-
-
-
0.211**
US
Change in Nonfarm Payrolls
-
-3.677***
-
-
-
-
0.132**
1.431***
0.00
-
-0.52**
-
US
Durables Ex Transportation
-
-
-
-0.073***
-0.261***
0.00
-
-
-
0.206***
-
-
Notes: Table 7 presents the estimation of scheduled and unscheduled macro news effect that over different regimes of business in our sample period, 𝜃𝑞 in equation (8). CN presents
the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain.*, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and
recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession.
69
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 7. Estimation Results of News Surprise (continued)
VOLAT
CN
US
US
US
Scheduled News
Univ. of Michigan Confidence - Preliminary
Wholesale Inventories MoM
Net Long-term TIC Flows
Recession
-
DEPTH
Expansion
-
P-diff
QSPREAD
Recession
Expansion
-
0.189***
-
-
0.862***
-
-0.167***
-
P-diff
Recession
-
0.11*
-
-0.14**
-
0.597***
-
-
SLOPE
Expansion
P-diff
Recession
Expansion
P-diff
-
-
-0.293***
-
-
-
-
-
-
-
0.591***
0.03
-
-
-
-
0.342***
-
-
-
-
-
US
New Home Sales
-
-
-
-0.458***
US
PCE Core MoM
-
-2.244***
-
-0.717***
-
-
0.197**
1.484***
0.00
-
-
-
US
Personal Consumption - Preliminary
-
-
-
-0.293***
-
-
-
-1.051***
-
0.538*
-
-
US
Pending Home Sales MoM
-
-
-
0.419***
-
-
0.928/***
-
-
-
-
-
-
1.799**
-
0.352***
-
-
-0.187**
-
-
-
-
-
-
-0.168*
-
-
-
-
-
-
-
-
-
-
-0.386***
-
-
-
-
-
-
-
-
0.2288**
-
-
-
-
-
-
-
-
-
-0.082*
-
-
-0.359**
-
-
-
-
-
-
0.277**
-
-
Panel C: European Countries
GE
GE
GE
GE
GE
IFO Business Climate
Imports QoQ
Industrial Production SA MoM – Preliminary
Private Consumption QoQ
Retail Sales MoM
-
-
-1.282**
-
-
-
0.345**
-
-
-
GE
ZEW Survey Current Situation
-
-4.116**
GE
Exports QoQ
-
-
0.137*
-
-
-
-
-
-
-
-
-
GE
ZEW Survey Expectations
-
-
-
-
-
-
-
-
-
-0.607***
-
-
GE
Factory Orders WDA YoY - Preliminary
-
1.268*
-
-
-
-
-
-
-
-
-
-
Notes: Table 7 presents the estimation of scheduled and unscheduled macro news effect that over different regimes of business in our sample period, 𝜃𝑞 in equation (8). CN presents
the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain.*, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and
recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession.
70
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 7. Estimation Results of News Surprise (continued)
VOLAT
CN
Scheduled News
Recession
DEPTH
Expansion
P-diff
Recession
QSPREAD
Expansion
P-diff
Recession
SLOPE
Expansion
P-diff
Recession
Expansion
P-diff
-
-
GE
Unemployment Rate
-
-
-
-
-
-
-
-
-
0.194*
GE
Construction Investment QoQ
-
-
-
-
-
-
-
-
-
-
-
-
0.293*
0.17
-
-
-
-
-
GE
PPI MoM
-
-
-
-
-
0.182***
-
-
-0.143**
-
-0.259**
-
-
-0.259**
-
IT
Business Confidence
-
-1.426*
IT
GDP WDA QoQ - Preliminary
-
-
-
-
-0.519***
-
-
-
-
-
-
-
IT
Retail Sales MoM
-
-
-
-
-0.183**
-
-0.155**
-
-
-
-
-
IT
Total investments
-
-
-
-
0.398**
-
-
-
-
-
-
-
IT
Trade Balance Total
-
-1.657*
-
-0.102*
-
-
-
-
-
-
-
-
-
-1.387**
-
-
-0.159*
-
-
-
-
-
-
-
-
4.676**
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
0.00
-0.109***
0.088***
0.00
0.00
-0.143***
0.098***
0.00
SP
SP
PO
CPI MoM
Retail Sales WDA YoY
CPI MoM
-
-
-
-
-
-
0.593**
0.05***
-0.131***
0.01
-0.007***
0.044***
0.00
0.015***
-0.041***
-
0.204**
-
-0.017***
0.038***
0.00
0.017***
-0.026**
Panel D: Unscheduled News
US
EC
Unscheduled News
Unscheduled News
Notes: Table 7 presents the estimation of scheduled and unscheduled macro news effect that over different regimes of business in our sample period, 𝜃𝑞 in equation (8). CN presents
the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain.*, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion
and recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession.
71
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 8. Estimation Results of Pure News
VOLAT
CN
Scheduled News
DEPTH
QSPREAD
P-diff
Recession
SLOPE
Recession
Expansion
P-diff
Recession
Expansion
Expansion
P-diff
Recession
Expansion
P-diff
-
-
-
0.222***
0.19**
0.04
-
-
-
-0.434***
-
-
1.833***
-
-
0.308***
0.47***
0.00
-0.244**
-
-
0.849***
-
-
-
-
-
0.32***
0.219**
0.52
-0.238**
-
-
-
-
-
1.932***
1.621**
0.37
0.775***
1.357***
0.00
-0.919***
-0.743***
0.02
0.33**
-
-
-
-
-
0.268***
0.557***
0.00
-
-
-
-
-
-
-
0.127***
0.187***
0.12
-
-
-
-
-
-
-
0.215***
0.236***
0.12
-
-
-
-
-
-
-
0.648***
-
-
-
-
-
-
-
-
-
***
-
-
-
-
-
-
-
-
0.146***
0.527***
0.00
-0.13*
-
-
-
-
-
-
0.115**
0.472***
0.00
-0.19***
-
-
-
-
-
-
-
-
-
-
-
-
Panel A: Euro Zone Macro News
EC
Business Climate Indicator
EC
CPI Core YoY - Final
EC
CPI Estimate YoY
EC
ECB Announces Interest Rates
EC
Govt Expend QoQ - Preliminary
EC
Industrial New Orders SA (MoM)
EC
EC
EC
EC
EC
EC
Industrial Production SA MoM
Labour Costs YoY
PMI Manufacturing - Preliminary
Retail Sales MoM
Trade Balance SA
ZEW Survey Expectations
-
-
-
-
1.21
*
1.666***
-
-
-
0.553
-
-
-
-
0.191**
Panel B: US Macro News
US
ADP Employment Change
3.332***
1.956**
0.85
0.796***
1.763***
0.00
-0.493***
-0.55***
0.05
-
-
-
US
Avg Hourly Earning MOM Prod
4.445***
5.528***
0.00
0.438***
0.886***
0.00
-1.739***
-2.769***
0.00
0.297**
0.894***
0.00
-
0.281***
0.718***
0.00
-
-0.999***
-
-
-
-
-
0.617***
1.000***
0.00
-0.985***
-0.371***
0.02
-
-
-
-
0.456***
1.134***
0.00
-0.571***
-0.812***
0.00
0.569**
-
-
US
US
US
Construction Spending MoM
-
Consumer Confidence Index
1.757***
Core PCE QoQ - Preliminary
2.243***
-
Notes: Table 8 presents estimation results of the selected pure news effect and unscheduled news effect over different regimes of business in our sample period, 𝜉𝑞 in equation (9). CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜉𝑞′ in equation (9), which is the difference between the coefficients in expansion and
recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession
72
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 8. Estimation Results of Pure News (continued)
VOLAT
CN
US
US
Scheduled News
ISM Milwaukee
ISM Non-Manf. Composite
Recession
DEPTH
Expansion
P-diff
QSPREAD
SLOPE
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
-
0.202**
1.462***
0.00
-
-
-
-
-
-
3.11***
-
-
-
-
-
-
-
-
3.952**
-
0.462***
0.00
-0.915***
-
0.325***
0.378***
0.00
-
-
-
-
-
-
US
NAHB Housing Market Index
-
1.529**
US
New Home Sales
-
4.012***
-
0.628***
1.355***
0.00
-1.285***
-1.01***
0.00
-
0.453**
-
US
Nonfarm Productivity - Final
-
-
-
0.278***
0.87***
0.00
-0.253*
-0.308*
0.4
-
-
-
US
Empire Manufacturing
-
-
-
0.466***
0.801***
0.00
-0.467***
-0.317***
0.51
-
-
-
US
Existing Home Sales
1.606***
2.119***
0.00
0.492***
1.153***
0.00
-
-
-
-
-
-
Factory Orders
0.947***
1.721***
0.00
0.545***
1.147***
0.00
-0.196***
-0.452***
0.00
-
-
-
FOMC Rate Decision
5.078***
4.782***
0.07
0.244***
0.873***
0.00
-1.977***
-1.334***
0.01
-
-
-
GDP Annualized QoQ - Advance
5.25**
4.343*
0.67
1.083***
0.968***
0.12
-2.544***
-0.986**
0.55
-
-
-
Housing Starts
1.105*
1.797*
0.05
0.333***
0.836***
0.00
-
-0.316**
-
-
-
-
-
0.302***
-
-
-
-
-
-
-
-
-
0.506***
-
-
-0.362***
-
-
-
-
-
0.852***
1.292***
0.00
-0.464***
-0.297***
0.63
-
-
-
-
0.451***
0.963***
0.00
-0.501***
-0.498***
0.00
-
-
-
0.697***
0.00
-0.156**
-
-
-
-
-
US
US
US
US
US
US
US
US
IBD/TIPP Economic Optimism
Import Price Index MoM
Industrial Production MoM
Initial Jobless Claims
-
-
-
US
Business Inventories
-
-
-
0.596***
US
Chicago Purchasing Manager
-
-
-
0.517***
0.969***
0.00
-
-
-
0.309**
-0.33*
0.03
US
Net Long-term TIC Flows
1.365**
2.905***
0.02
0.73***
1.391***
0.00
-1.039***
-0.484***
0.82
0.232*
-
-
US
ISM Manufacturing
-
-
-
0.584***
0.407***
0.03
-0.762***
0.446**
0.00
-
-
-
Notes: Table 8 presents estimation results of the selected pure news effect and unscheduled news effect over different regimes of business in our sample period, 𝜉𝑞 in equation (9). CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜉𝑞′ in equation (9), which is the difference between the coefficients in expansion and
recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession
73
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 8. Estimation Results of Pure News (continued)
VOLAT
CN
US
US
US
Scheduled News
Nonfarm Productivity - Preliminary
PCE Core MoM
Core PCE QoQ - Advance
Recession
DEPTH
Expansion
2.361**
-
-
2.69***
-
-
P-diff
QSPREAD
Recession
Expansion
-
0.614***
0.265***
-
0.477***
1.427***
-
-0.541**
0.531***
SLOPE
P-diff
Recession
0.00
-0.619***
Expansion
P-diff
-
-
0.00
-0.725***
-1.285***
-
2.044***
0.00
0.269***
Recession
Expansion
P-diff
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
US
Pending Home Sales MoM
-
-
-
0.571***
US
Philadelphia Fed Business Outlook
-
2.265***
-
0.605***
1.508***
0.00
-0.804***
-0.213***
0.1
-
-
-
US
PPI Ex Food and Energy MoM
-
-
-
0.304***
1.305***
0.00
-1.322***
-0.372***
0.03
-
-
-
US
Retail Sales Ex Auto MoM
2.254***
2.089**
0.03
0.309***
0.918***
0.00
-0.897***
-1.061***
0.00
-
-
-
US
Trade Balance
1.143**
2.76***
0.05
0.87***
0.803***
0.00
-1.312***
-1.048***
0.00
0.292*
-
-
1.1**
1.559**
0.29
0.214***
0.573***
0.00
-0.368***
-0.315***
0.23
-
-
-
-
1.42*
-
0.44***
0.632***
0.00
-0.183**
-
-
-
-
-
0.905***
-1.161***
US
US
US
US
Univ. Michigan Confidence - Preliminary
Wholesale Inventories MoM
CPI Ex Food and Energy MoM
2.134***
-
-
0.547***
0.00
-1.736***
0.03
-
-
-
Durables Ex Transportation
1.792***
2.366***
0.16
0.568***
0.825***
0.00
-1.04***
-1.084***
0.00
-
-
-
-
-
-
-
0.168*
-
-
-
-
-
-
-
Panel C: European Countries Macro News
GE
Construction Investment QoQ
GE
Factory Orders WDA YoY - Preliminary
1.598***
-
-
-
0.219***
-
-
-
-
-
-
-
GE
GDP SA QoQ - Preliminary
2.292***
-
-
-
0.277***
-
-
-
-
-
-
-
GE
IFO Business Climate
2.467***
2.864***
0.06
0.436***
0.687***
0.00
-0.214***
-0.285***
0.1
-
-
-
GE
PPI MoM
1.152**
-
-
0.11**
0.232***
0.03
-
-
-
-
-
-
Retail Sales MoM
1.15**
-
0.226***
-
-0.144**
-
-
-
-
-
GE
-
-
Notes: Table 8 presents estimation results of the selected pure news effect and unscheduled news effect over different regimes of business in our sample period, 𝜉𝑞 in equation (9). CN presents the
corresponding country name of the news: EC - Euro Zone Aggregate; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜉𝑞′ in equation (9), which is the difference between the coefficients in expansion and recession. A
significant value of P_diff indicate that the coefficients are statistically different over expansion and recession
74
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 8. Estimation Results of Pure News (continued)
VOLAT
CN
GE
GE
GE
Scheduled News
Industrial Production SA MoM Preliminary
Unemployment Rate
ZEW Survey Expectations
Recession
1.233**
-
DEPTH
Expansion
-
P-diff
SLOPE
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
-
0.18***
0.408***
0.00
-
-
-
-
-
-
-
-0.103**
-
-
-
-
-
-
-
-
-
-1.095**
-
-1.367*
-
-
-
-
-
-
-
-
-
-
-
-
-
Recession
QSPREAD
GE
ZEW Survey Current Situation
-
-
-
-
1.756***
FR
PPI MoM
-
-
-
-
-
-
0.147*
-
-
-
-
-
SP
CPI EU Harmonised YoY - Final
-
-
-
0.154***
-0.159**
0.01
-
-
-
-
-
-
PO
GDP YoY - Final
-
-
-
-
-
-
-
-
-
-
1.111**
-
PO
CPI MoM
-
-
-
-
0.186**
-
-0.234***
-
-
-
-
-
-
0.105**
0.206***
0.04
-
-
-
-0.385***
-
-
-
-
0.315**
-
-
-
-
-
-
-
-
0.185***
0.136**
0.64
-
-
-
-
-
-
-
0.099**
0.244***
0.02
-
-
-
-
-
-
0.172**
0.21
-
-
-
-
-
-
IT
IT
IT
IT
IT
Business Confidence
GDP WDA QoQ - Preliminary
Industrial Production WDA YoY
Retail Sales MoM
Trade Balance Total
-
-
-
-
-
0.108**
Panel D: Unscheduled News
US
Unscheduled News
0.054***
-0.152***
0.00
-0.005***
0.035***
0.00
0.013***
-0.032***
0.00
-0.109***
0.086***
0.00
EC
Unscheduled News
-
0.188**
-
-0.014***
0.025***
0.00
0.014***
-0.019*
0.02
-0.143***
0.098***
0.00
Notes: Table 8 presents estimation results of the selected pure news effect and unscheduled news effect over different regimes of business in our sample period, 𝜉𝑞 in equation (9). CN presents the
corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news
are at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜉𝑞′ in equation (9), which is the difference between the coefficients in expansion and recession. A significant value of P_diff
indicate that the coefficients are statistically different over expansion and recession
75
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 9. Estimation Results of Good and Bad News
VOLAT
DEPTH
Recession
Expansion
Aggregated Good
1.168***
1.188***
Aggregated Bad
0.879***
1.218***
US unscheduled
0.054***
-0.145***
EC unscheduled
-
0.194**
QSPREAD
P-diff
Recession
Expansion
0.00
0.374***
0.744***
0.00
0.329***
0.714***
0.00
-0.006***
0.038***
-
0.014***
0.03***
SLOPE
P-diff
Recession
Expansion
0.00
-0.421***
-0.354***
0.00
-0.408***
-0.42***
0.00
0.013***
-0.035***
0.00
0.015***
-0.021**
P-diff
Recession
Expansion
P-diff
0.00
0.103***
-
-
0.00
-
0.098**
-
0.00
-0.109***
0.087***
0.00
0.01
-0.142***
0.098***
0.00
Notes: Table 9 presents the estimation results of aggregated good and bad news effect and unscheduled news effect in different regimes in equation 𝜌𝑔 and 𝜌𝑏 (10). *, **, *** denotes the prob.
of insignificance of news are at 1%, 5% and 10% levels respectively. Quoted Spread is denoted as QSpread, which is the percentage of the price difference between the best bid and the best ask
price accounted for the mid-price of the LOB. Volat is the filtered absolute value of 5-min return which is calculated with size-weighted price. Depth is the sum of the product of size and price
in each interval. Slope measures the elasticity of the LOB supply and demand curve. P_diff presents the P value of 𝜌𝑔′ and 𝜌𝑏′ in equation (10), which is the difference between the coefficients
in expansion and recession. A significant value of P_diff indicate that the coefficients are statistically different over expansion and recession.
76
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 10. Robustness Results of Surprise on Alternative Slopes
SLOPE
CN
Scheduled News
NORMSLOPE
Recession
Expansion
P-diff
Recession
WSLOPE
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone Macro News
EC
Labour Costs YoY
-
-
-
-
1.02**
-
-
-
-
EC
Retail Sales MoM
-
-
-
-
-0.649***
-
-
-
-
EC
CPI Core YoY - Final
-0.271**
-
-
-
-
-
-
-
-
EC
CPI Estimate YoY
0.832***
-
-
-
-
-
-
-
-
EC
Business Climate Indicator
0.21*
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
0.92
-
-
-
-
-
-0.876**
-
0.88
-
-
-
-
-
-
-
0.72
-
-
-
EC
EC
Trade Balance SA
ZEW Survey Expectations
0.233
**
0.276
**
*
-
-
0.176
-0.771***
-
-0.304***
-
***
-0.317
*
Panel B: US Macro News
US
US
US
Avg Weekly Hours Production
Business Inventories
Chicago Purchasing Manager
US
Construction Spending MoM
US
Core PCE QoQ - Preliminary
US
US
US
US
US
-
-
-
-0.274
-
0.325
0.32
*
**
**
-
-
-
0.188
-0.631***
-
-
-
-1.326***
-
-0.705*
-
-
CPI Ex Food and Energy MoM
-
-
-
-
0.515**
-
-0.401**
-
-
Factory Orders
-
-
-
-
-0.27*
-
-
-
-
0.02
-
-
-
-
-
-
-
-
-
-
-
GDP Annualized QoQ - Preliminary
Import Price Index MoM
Initial Jobless Claims
-
-
-
0.572
**
0.214
**
-
0.391
*
-0.939
***
-
-0.319
***
Notes: Table 10 presents the estimation results of significant scheduled and unscheduled news effect on slope, NORMSLOPE and WSLOPE. CN presents the corresponding country
name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news are at
1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and recession. A significant
value of P_diff indicate that the coefficients are statistically different over expansion and recession.
77
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 10. Robustness Results of Surprise on Alternative Slopes (continued)
SLOPE
CN
US
US
US
Scheduled News
Nonfarm Productivity - Preliminary
PCE Core MoM
Personal Consumption - Preliminary
NORMSLOPE
Recession
-
Expansion
-
-
0.538
Recession
-
*
P-diff
0.466
Expansion
***
-
-0.465
**
P-diff
Recession
Expansion
P-diff
***
0.62
-
-
-
***
-
-
-
-
**
0.03
-
-
-
***
-
-
-
-
0.389
-0.59
-
-
WSLOPE
0.705
US
Personal Spending
-
-
-
US
PPI Ex Food and Energy MoM
-
-
-
0.207*
-0.319**
0.06
-
-
-
US
Retail Sales Ex Auto MoM
-
-
-
0.177**
0.262*
0.44
0.294*
-
-
US
Durables Ex Transportation
0.206*
-
-
-
-
-
-
-
-
US
Existing Home Sales
-
0.942***
-
-
-
-
-
US
US
US
US
FOMC Rate Decision
Unemployment Rate
Change in Nonfarm Payrolls
Univ. Michigan Confidence Preliminary
-
0.561
-0.293
GDP Annualized QoQ - Advance
-
US
ADP Employment Change
-
US
US
ISM Manufacturing
ISM Milwaukee
ISM Non-Manf. Composite
0.273
**
**
-
0.3
-
-0.52
-
US
US
***
-
-
-
-
-
0.253
-
-0.772
-
-
-
-
-
-
-
-
-
-
-
-
-
-
0.369
-
0.212
*
-
-0.747
-
0.507
*
-
2.212
*
0.82
-
***
-
-
-
-
-
-
-
-
-
***
-
-
-
-
**
-
-
-
-
***
0.00
-
-
-
0.797
-
***
-
-
-
-
-
-
-
-
**
-
-
**
**
0.638
-2.028
Notes: Table 10 presents the estimation results of significant scheduled and unscheduled news effect on slope, NORMSLOPE and WSLOPE. CN presents the corresponding country
name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news are at 1%,
5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and recession. A significant value of
P_diff indicates that the coefficients are statistically different over expansion and recession.
78
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 10. Robustness Results of Surprise on Alternative Slopes (continued)
SLOPE
CN
Scheduled News
NORMSLOPE
Recession
Expansion
P-diff
Recession
WSLOPE
Expansion
P-diff
Recession
Expansion
P-diff
Panel C: European Countries Macro News
GE
IFO Business Climate
GE
PPI MoM
GE
Retail Sales MoM
GE
GE
GE
-
-
-
***
-
-
-
0.971***
-
-
-
-
-
-
-
-
-
-0.19
-0.359**
-
-
0.191*
-0.33**
0.27
-
-
-
Unemployment Rate
0.194*
-
-
-
-
-
-
-
-
ZEW Survey Current Situation
0.277**
-
-
-
-
-
-
-
-
-0.607
**
-
-
-
-
-
-
-
-
-0.259
**
-
-
-
-
-
-
-
-
0.00
0.038***
0.148***
0.00
0.118***
0.072***
0.00
0.00
***
-
-
-
-
-
ZEW Survey Expectations
IT
Business Confidence
CN
Unscheduled News
US
Unscheduled News
-0.109***
0.088***
Unscheduled News
***
***
EC
-
-0.143
0.098
-0.018
Notes: Table 10 presents the estimation results of significant scheduled and unscheduled news effect on slope, NORMSLOPE and WSLOPE. CN presents the corresponding country name
of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news are at 1%, 5% and
10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the coefficients in expansion and recession. A significant value of P_diff
indicates that the coefficients are statistically different over expansion and recession.
79
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 11. Robustness Results of Surprise on Alternative Volatilities
VOLAT
CN
Scheduled News
Y
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone Macro News
EC
Govt Expend QoQ - Preliminary
-3.965*
-
-
-
-
-
EC
Labour Costs YoY
-1.337*
4.308*
0.05
-
6.76**
-
EC
Retail Sales MoM
-1.488**
-
-
-
-
-
-
-3.459**
-
-
-
-
***
Panel B: US Macro News
US
US
US
US
US
US
US
ADP Employment Change
Change in Nonfarm Payrolls
Consumer Confidence Index
Empire Manufacturing
Factory Orders
FOMC Rate Decision
GDP Annualized QoQ - Advance
US
GDP Annualized QoQ – Preliminary
US
Housing Starts
US
Initial Jobless Claims
US
ISM Non-Manf. Composite
US
US
US
US
US
US
US
US
US
NAHB Housing Market Index
Nonfarm Productivity - Final
Nonfarm Productivity - Preliminary
PCE Core MoM
Personal Consumption - Preliminary
PPI Ex Food and Energy MoM
Trade Balance
Import Price Index MoM
Core PCE QoQ - Advance
-3.677
*
*
**
-0.688
0.908
1.112
0.775
**
-
-
-3.165
***
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-3.673
***
-3.881
***
-
-
-
-
1.671
**
-
-4.041
***
-
*
-
-
-
-3.641
-1.462***
-
-
-
-
-
-
-1.294***
-
-
-1.83***
-
1.58**
-6.649***
0.00
-
-
-
1.975
1.39
***
**
1.542
*
-
-
-3.725
0.02
1.256
*
0.61
***
-
-1.605
*
0.00
**
-2.224
-
0.858
-1.961
***
**
-
-
-2.1
-
-
-
-
1.091
*
**
-0.888
-
-
*
-
-
-2.276
**
1.93
**
-
-
-
-8.989
-
-1.616
-
**
**
-
-
-
-
-
-
2.191
**
0.01
*
-
-4.516
Notes: Table 11 presents the estimation results of surprise of the selected significant scheduled and unscheduled macro news
effect of two alternatives of volatility. CN presents the corresponding country name of the news: EC - Euro Zone; GE - German;
US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news are at
1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the
coefficients in expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over
expansion and recession.
80
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 11. Robustness Results of Surprise on Alternative Volatilities (continued)
VOLAT
CN
Scheduled News
Recession
Y
Expansion
P-diff
Recession
Expansion
P-diff
Panel C: European Countries Macro News
GE
Retail Sales MoM
-1.282**
-
-
-
-
-
-
-
-
-
GE
ZEW Survey Expectations
-
GE
Factory Orders WDA YoY - Preliminary
-
1.268*
-
-
-
-
GE
IFO Business Climate
-
1.779**
-
-
2.056***
-
SP
CPI MoM
-
-1.387**
-
-
-1.405**
-
-
**
-
-
**
-
-1.426
*
-
-
-
-1.657
*
-
-
-
SP
IT
IT
Retail Sales WDA YoY
Business Confidence
Trade Balance Total
IT
Retail Sales MoM
CN
Uncheduled News
US
EC
-
-4.115
**
4.676
5.143
-1.712
**
-
-
-
-
-
Unscheduled News
0.05***
-0.131***
0.02
0.097***
-
-
Unscheduled News
-
0.204**
-
0.124**
0.191**
0.00
Notes: Table 11 presents the estimation results of surprise of the selected significant scheduled and unscheduled macro news
effect of two alternatives of volatility. CN presents the corresponding country name of the news: EC - Euro Zone; GE - German;
US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of insignificance of news are
at 1%, 5% and 10% levels respectively. P_diff presents the P value of 𝜃𝑞′ in equation (8), which is the difference between the
coefficients in expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over
expansion and recession.
81
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 12. Number of Significant News
VOLAT
Country
Total
DEPTH
QSPREAD
SLOPE
Recession
Expansion
SD
Recession
Expansion
SD
Recession
Expansion
SD
Recession
Expansion
SD
Panel A: Surprise
EC
16
3
1
1
7
11
2
7
11
0
5
0
0
FR
3
0
0
0
0
0
0
0
0
0
0
0
0
GE
13
1
3
0
4
3
0
4
4
0
4
0
0
IT
8
0
2
0
1
4
0
1
4
0
1
0
0
PO
2
0
0
0
0
0
0
0
0
0
0
0
0
SP
4
0
2
0
0
1
0
1
0
0
0
0
0
US
43
11
12
2
28
32
19
28
34
12
9
3
0
Total
89
15 (17%)
19 (21%)
3 (3%)
40 (45%)
51 (57%)
21 (24%)
41 (46%)
53 (60%)
12 (13%)
19 (21%)
3 (3%)
0 (0%)
Panel B: Pure News
EC
14
4
1
0
9
12
6
5
1
1
3
0
0
FR
3
0
0
0
1
1
0
1
0
0
0
0
0
GE
10
6
1
1
4
9
3
1
1
0
0
0
0
IT
7
0
0
0
4
5
1
0
0
0
1
0
0
PO
2
0
0
0
0
0
0
1
0
0
0
1
0
SP
2
0
0
0
1
1
1
0
0
0
0
0
0
US
36
17
19
8
34
35
32
29
25
14
5
4
2
Total
74
27 (30%)
21 (24%)
9 (10%)
53 (60%)
63 (71%)
42 (47%)
37 (42%)
27 (30%)
15 (17%)
9 (10%)
5 (6%)
2 (2%)
Notes: Table 12 shows the percentage of significant news in each country for estimation results of VAR-STR with news surprise and pure news. Country provide the countries name corresponding to the
news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. Panel A and B contains the significant and state-dependent news information surprise and pure
news respectively. The content in the table, that is, (%) stands for the percentage of the number of significant dependent news in regression and expansion. Total is the sum of significant news of each
characteristics in two regimes.SD stands for the number of state dependent news.
82
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 13. Number of Significant News in Robustness
EC
FR
GE
IT
PO
SP
US
Total
Robustness Check of Alternative Slopes
SLOPE
NORMSLOPE
WSLOPE
Recession
5
0
4
1
0
0
9
19 (21%)
Expansion
0
0
0
0
0
0
3
3 (3%)
Sum
5
0
4
1
0
0
12
22 (25%)
Recession
1
0
1
1
0
0
11
14 (16%)
Expansion
3
0
0
1
0
0
20
24 (27%)
Sum
4
0
1
2
0
0
31
39 (44%)
Recession
0
0
0
1
1
0
6
8 (9%)
Expansion
0
0
1
0
0
0
4
5 (6%)
Sum
0
0
1
1
1
0
10
13 (15%)
Robustness Check of Alternative Volatilities
VOLAT
Y
Recession
3
0
1
0
0
0
11
15 (17%)
Expansion
1
0
3
2
0
2
12
20 (22%)
Sum
4
0
4
5
0
5
23
41 (46%)
Recession
5
0
4
1
0
0
9
19 (21%)
Expansion
0
0
0
0
0
0
3
3 (3%)
Sum
5
0
4
1
0
0
12
22(25%)
N
16
3
13
8
2
4
43
89
Notes: Table 13 shows the percentage of significant news surprise category in each country of alternative methods of
slope and volatility. Country provide the countries name corresponding to the news: EC- Euro Zone, FR-France,
GE-German, IT-Italy, PO-Poland, SP-Spain, and US-United States. Panel A is the significant news information of
three slope measures. Panel B is the significant news information of two return volatility measures. The content in
the table, that is, (%) stands for the percentage of the number of significant dependent news in regression and
expansion. N is the total of news category in each country. Sum in every section of characteristic is the sum of
significant news regardless regime. Total is the sum of significant news of each characteristics in two regimes.
83
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 14. Robustness Results of Surprise on Depth at Ask and Bid Sides
Depth
CN
Ask Side
Scheduled News
Bid Side
Recession
Expansion
P-diff
Recession
Expansion
P-diff
-
0.284***
-
-
0.277***
-
-
***
-
-
Panel A: Euro Zone News
EC
Business Climate Indicator
EC
CPI Core YoY - Final
EC
CPI Estimate YoY
EC
GDP SA QoQ - Final
EC
Govt Expend QoQ - Preliminary
EC
EC
EC
EC
EC
EC
EC
EC
EC
Gross Fix Cap QoQ - Final
Gross Fix Cap QoQ - Preliminary
Household Cons QoQ - Preliminary
Labour Costs YoY
Retail Sales MoM
Trade Balance SA
ZEW Survey Expectations
Industrial New Orders SA (MoM)
Industrial Production SA MoM
-0.243
-
-0.097
*
-0.095*
0.214*
0.55
-
-
-
-
-0.249**
-
-
-
-
0.634***
-
-
0.688***
-
-
-
**
-
-
***
-
0.656
***
-0.611
***
-
0.367
**
0.346
***
-0.551
-0.093
*
-0.087
*
0.136
-0.533
***
***
-
**
0.288
0.3
0.01
-
***
-0.292
***
-
0.04
0.00
-
0.229
0.719
-0.648
***
-
*
0.00
*
-
***
-
***
-
***
-
*
-
0.206
-0.420
-
-0.216
-
0.242
-
-0.263
-
0.158
-
-
-
-0.117
-
-1.626***
-
-
*
-
-
-1.566***
-
-
-
Panel B: US News
US
US
US
US
US
US
US
US
US
US
ADP Employment Change
Avg Weekly Hours Production
Business Inventories
Change in Nonfarm Payrolls
Chicago Purchasing Manager
Construction Spending MoM
Consumer Confidence Index
Core PCE QoQ - Advance
Core PCE QoQ - Preliminary
Durables Ex Transportation
-0.242
***
-0.131
***
0.144
0.150
0.374
US
Existing Home Sales
US
Factory Orders
GDP Annualized QoQ - Advance
***
-0.078
0.220
US
**
-
Empire Manufacturing
FOMC Rate Decision
***
0.078
US
US
***
**
***
0.196
0.502
**
***
-0.472
***
0.593
***
0.855
***
0.446
***
-1.456
***
-2.271
***
-0.227
***
0.09
0.01
0.04
0.00
0.00
0.03
-
0.116**
0.934***
0.00
-0.087**
-
-
0.284
0.130
***
-
-0.117
0.101
**
**
0.256
***
0.00
-
**
-0.226
***
-
0.314
***
0.267
***
***
0.00
***
0.55
0.740
***
-
0.589
***
0.01
0.624
***
-
**
0.04
-2.335
***
-
-0.310
***
-
0.232
-0.429
-1.143
-
-
0.176***
0.669***
0.00
-0.090**
0.154**
0.36
-
-
0.060
0.214
*
**
-0.429
***
0.01
Notes: Table 14 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news
on Depth at ask side and bid side with all levels in LOB considered. CN presents the corresponding country name of the news: EC
- Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
84
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 14. Robustness Results of Surprise on Depth at Ask and Bid Sides (continued)
Depth
CN
US
US
US
Ask Side
Scheduled News
GDP Annualized QoQ - Preliminary
Housing Starts
Import Price Index MoM
Recession
0.348
***
-0.121
**
-
Bid Side
Expansion
-0.906
***
P-diff
0.00
Recession
0.435
***
Expansion
0.03
0.394
***
-
0.438
***
0.04
***
-
-0.912
0.254
***
0.474
***
***
-
-
0.462
0.00
-
0.076
*
P-diff
***
US
Industrial Production MoM
-
0.419
US
Initial Jobless Claims
-
-0.525***
-
-
-0.599***
-
US
ISM Manufacturing
-0.098**
0.359***
0.00
-
0.434***
-
US
ISM Milwaukee
-
0.824***
-
-
1.110***
-
US
US
US
US
US
US
US
US
US
ISM Non-Manf. Composite
NAHB Housing Market Index
Net Long-term TIC Flows
New Home Sales
Nonfarm Productivity - Final
Nonfarm Productivity - Preliminary
PCE Core MoM
Pending Home Sales MoM
Personal Consumption - Preliminary
0.269
***
-0.201
-0.411
***
***
0.099
0.504
*
***
0.444
***
-0.219
**
***
-1.926
***
-0.286
***
-
-
-
-
-0.818
***
0.160
**
-0.602
***
-
0.09
0.78
-
-
-
0.394
-
-2.214
***
0.00
-0.200
***
-
-0.196
***
-
-0.509
***
-
-
***
-
***
0.98
***
-
-0.678
0.384
***
-
0.203
-0.776
***
-
-0.331
***
-
**
0.362
***
0.02
0.155
***
0.00
Personal Spending
0.212
US
Philadelphia Fed Business Outlook
-0.084*
0.339***
0.01
-0.092*
0.363***
0.00
US
PPI Ex Food and Energy MoM
0.441***
-0.640***
0.00
0.486***
-0.684***
0.01
US
PPI MoM
-0.192***
0.950***
0.01
-0.172***
1.087***
0.00
***
*
0.01
***
***
0.00
***
0.01
*
-
0.661
***
0.05
0.489
***
0.27
US
US
US
US
Trade Balance
Unemployment Rate
Univ. of Michigan Confidence - Preliminary
Wholesale Inventories MoM
CPI Ex Food and Energy MoM
0.171
-0.144
-
***
-0.148
0.415
***
0.235
***
0.946
***
-
0.00
-
0.211
-0.142
**
-
-0.398
0.418
0.124
0.084
-0.123
0.369
-
US
US
0.250
0.00
***
*
***
Notes: Table 14 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news
on Depth at ask side and bid side with all levels in LOB considered. CN presents the corresponding country name of the news:
EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob.
of insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
85
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 14. Robustness Results of Surprise on Depth at Ask and Bid Sides (continued)
Depth
CN
Ask Side
Scheduled News
Bid Side
Recession
Expansion
P-diff
Recession
Expansion
P-diff
-
0.300***
-
0.122**
0.319***
0.44
-
-0.167
*
-
-
Panel C: European Countries
GE
IFO Business Climate
GE
Imports QoQ
GE
Industrial Production SA MoM - Preliminary
GE
Private Consumption QoQ
GE
IT
IT
IT
IT
IT
IT
SP
FR
ZEW Survey Current Situation
Business Confidence
GDP WDA QoQ - Final
GDP WDA QoQ - Preliminary
Total investments
Trade Balance Total
Retail Sales MoM
Unemployment Rate
PPI MoM
-0.150
*
-
-0.377***
-
-
-0.321***
-
0.217**
-
-
0.229**
-0.296**
-
***
-
-
***
-
-
***
-
-
-
0.395
*
-
-
-
-0.409
**
-
-
***
-
-
-
-
-
-
0.304
-
-0.226
-
0.784
-0.161
**
-
-
-
-
-
-
0.287
**
0.309
-
-0.601
-
***
-
-0.251
***
-
**
-
-
-
-0.398
-
-
-
0.00
-0.008***
0.032***
0.01
0.00
***
***
0.00
-
Panel D: Unscheduled News
US
EC
US Unscheduled News
-0.009***
0.031***
EC Unscheduled News
***
**
-0.015
-0.003
-0.017
0.021
Notes: Table 14 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news on
Depth at ask side and bid side with all levels in LOB considered. CN presents the corresponding country name of the news: EC Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
86
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 15. Robustness Results of Surprise on Slope at Ask and Bid Sides
Slope
CN
Ask Side
Scheduled News
Bid Side
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone News
EC
Business Climate Indicator
0.316*
-
-
0.344**
-
-
EC
CPI Core YoY - Final
-0.349*
-
-
0.365*
-
-
EC
CPI Estimate YoY
0.654***
-
-
-
-
-
EC
Govt Expend QoQ - Preliminary
-3.195***
-
-
-5.049***
-
-
-
***
-
-
***
-
-
-
-0.449
*
-
-
*
-
-
EC
EC
EC
EC
Gross Fix Cap QoQ - Preliminary
Household Cons QoQ - Preliminary
Industrial New Orders SA (MoM)
Industrial Production SA MoM
-1.165
2.486
*
***
-
-
-
-
-1.804
3.934
-
-
-
0.529
-
1.121***
-
-
1.038***
-
-
-
-
-
-
Panel B: US News
US
US
Existing Home Sales
ISM Manufacturing
0.373
**
US
Avg Weekly Hours Production
-
-
-
-
US
Core PCE QoQ - Preliminary
-
-
-
-0.754**
-
-
US
GDP Annualized QoQ - Preliminary
-
-
-
-0.810*
-
-
*
-
-
US
Personal Consumption - Preliminary
-0.836
**
-
-
-
0.825
-2.065***
-
-
-
-
Panel C: European Countries News
GE
GE
GE
GE
Construction Investment QoQ
Factory Orders WDA YoY - Preliminary
ZEW Survey Current Situation
ZEW Survey Expectations
-
-
-
-
-
-
-
-
*
-
-
**
-
-
**
-
-
-0.269
0.432
-
-
-
-0.774
US Unscheduled News
-0.116***
0.297***
0.00
-0.123***
0.225***
0.00
EC Unscheduled News
***
***
0.00
***
***
0.00
Panel D: Unscheduled News
US
EC
-0.193
0.150
-0.180
0.112
Notes: Table 15 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news on
Slope at ask side and bid side with all levels in LOB considered. CN presents the corresponding country name of the news: EC Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
87
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 16. Robustness Results of Surprise on Volatility at different levels in the LOB
Volatility
2nd to 5th level
CN
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Scheduled News
5th to 10th level
Panel A: Euro Zone News
EC
Govt Expend QoQ - Preliminary
-3.961*
-
-
-5.565*
-
-
EC
Labour Costs YoY
-1.472*
4.042*
0.06
-1.471**
3.579**
0.04
EC
Retail Sales MoM
-1.514*
-
-
-1.395**
-
-
EC
Gross Fix Cap QoQ - Preliminary
-
-
-
-3.402*
-
-
*
-
-
EC
Household Cons QoQ - Preliminary
-
-
-
4.041
-
-2.789*
-
-
Panel B: US News
US
US
US
US
US
ADP Employment Change
Change in Nonfarm Payrolls
Consumer Confidence Index
Core PCE QoQ - Advance
Empire Manufacturing
-
-3.428
-0.840
**
-
***
-
-
-8.217
0.948
*
-
*
-
-2.776*
-
-2.719
-0.859
**
-
***
-
-
-8.387
*
-
0.955
**
-
-
**
-
-
**
-
-
1.218
US
Existing Home Sales
1.255
US
Factory Orders
0.984**
-
-
0.967**
1.656**
0.08
US
FOMC Rate Decision
0.879***
-
-
2.506***
-
-
US
US
US
US
US
US
US
US
US
US
GDP Annualized QoQ - Advance
GDP Annualized QoQ - Preliminary
Housing Starts
Initial Jobless Claims
ISM Milwaukee
ISM Non-Manf. Composite
NAHB Housing Market Index
Nonfarm Productivity - Final
Nonfarm Productivity - Preliminary
PCE Core MoM
US
PPI Ex Food and Energy MoM
US
Trade Balance
US
Industrial Production MoM
US
Retail Sales Ex Auto MoM
-0.935
*
1.502
1.697
**
***
1.618
-3.375
***
-3.681
***
-
-
-
-0.975
-
-
-3.620
*
-
-
***
0.00
-
-1.575
**
*
1.433
**
-1.990
***
-1.586
**
*
-
-0.877
-6.169
-3.274
1.957
*
-3.461
**
-
0.00
0.31
-
1.749
*
-
-
**
0.00
-3.608
**
-
**
0.28
-1.189
**
-
-1.338
**
-
1.679
-
-
-
-
-
0.664*
-
-
-
-1.605
-
**
**
-
-
**
0.771*
-
0.07
-3.658
-3.952
-
-
-
**
-
1.539
***
1.152
*
-
Notes: Table 16 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news on
volatility at 2nd to 5th level and 5th to 10th level in the LOB considered. CN presents the corresponding country name of the news:
EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
88
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 16. Robustness Results of Surprise on Volatility at different levels in the LOB
(continued)
Volatility
2nd to 5th level
CN
Recession
Expansion
P-diff
Recession
Expansion
P-diff
-
1.811**
-
-
1.758**
-
-
-
Scheduled News
5th to 10th level
Panel C: European Countries
GE
IFO Business Climate
GE
Retail Sales MoM
GE
ZEW Survey Expectations
SP
CPI MoM
SP
IT
IT
Retail Sales WDA YoY
Trade Balance Total
Unemployment Rate Quarterly
-1.646
***
-
-
-
-4.622***
-
-
-4.702***
-
-
-1.289**
-
-
-1.282**
-
-
3.776
**
-
5.056
**
-
-2.354
***
-2.317
***
-
-
-
-1.557
**
-
-
-
-
-2.452
0.057**
-0.182***
0.00
-
*
-
-
Panel D: Unscheduled News
US
EC
US Unscheduled News
EC Unscheduled News
-
0.230
**
-
-
-0.204***
0.172
**
-
Notes: Table 16 presents the estimation results of news surprise effect of the significant scheduled and unscheduled macro news on
volatility at 2nd to 5th level and 5th to 10th level in the LOB considered. CN presents the corresponding country name of the news:
EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in
expansion and recession. A significant value of P_diff indicates that the coefficients are statistically different over expansion and
recession.
89
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB
Depth
Ask Side
2nd
CN
Scheduled News
to
5th
Bid Side
5th
level
Recession
Expansion
P-diff
to
10th
2nd to 5th level
level
Recession
Expansion
P-diff
Recession
5th to 10th level
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone Macro News
EC
Business Climate Indicator
-
0.624***
-
-
EC
CPI Estimate YoY
-
0.432*
-
-
EC
GDP SA QoQ - Final
-
-0.424*
-
-
EC
Gross Fix Cap QoQ Preliminary
0.656**
-0.702**
0.16
EC
Industrial New Orders SA (MoM)
-
0.432***
EC
Labour Costs YoY
-
EC
Trade Balance SA
-
0.416***
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-0.439*
-
-
-
-
-
-
-
0.993***
-
-
-
-
-
1.218***
-
-
-
-
0.28*
-
-
-
-
-
0.295*
-
-0.893**
-
-
-
-
-
-
-
-
-
-
0.42***
-
-
0.501***
-
-
-
-
-
-
-
-
-
-
-
-
EC
CPI Core YoY - Final
-
-
-
-
-0.251*
-
EC
Gross Fix Cap QoQ - Final
-
-
-
-0.815***
-
-
-
-
-
-
-
-
EC
Household Cons QoQ Preliminary
-
-
-
-0.826**
-
-
-
-
-
-0.959**
-
-
EC
Industrial Production SA MoM
-
-
-
-0.243**
-
-
-
-
-
-
-
-
EC
Govt Expend QoQ - Preliminary
-
-
-
-
-
-
-
-
-
1.104**
-
-
EC
PMI Manufacturing - Preliminary
-
-
-
-
-
-
-
-
-
0.247**
-
-
EC
Retail Sales MoM
-
-
-
-
-
-
-
-
-
-
-0.241*
-
EC
ZEW Survey Expectations
-
-
-
-
-
-
-
-
-
-
0.282*
-
Notes: Table 17 presents estimation results of the news surprise effect and unscheduled news effect on depth at
-0.196
**
2nd
to
5th
levels and
5th
to
10th
levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
90
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB (continued)
Depth
Ask Side
2nd
CN
Scheduled News
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
Recession
Expansion
P-diff
Recession
Expansion
-
-0.550**
-
-
-1.250***
-0.205**
-
-
-
-
0.312**
-
P-diff
5th to 10th level
Recession
Expansion
P-diff
Recession
Expansion
-
-0.196**
-1.435***
-
-
-1.434***
-
-
-
-
-
-0.176***
-0.289*
-
0.413***
-
-0.306***
0.724***
-
-
0.482***
P-diff
Panel B: US News
US
ADP Employment Change
US
Avg Weekly Hours Production
US
Business Inventories
US
Change in Nonfarm Payrolls
0.169*
-0.561***
0.03
-
-0.739***
-
-
-0.512**
-
-
-0.590***
US
Chicago Purchasing Manager
-
0.492***
-
-
0.603***
-
-
0.547***
-
-
0.775***
US
Construction Spending MoM
0.151*
0.365*
0.00
0.173*
0.544***
0.01
-
-
0.282***
0.522***
0.00
US
Consumer Confidence Index
-
0.920***
-
-
0.365**
-
-
1.054***
-
-
-
-
US
Core PCE QoQ - Advance
-
-1.922**
-
-
-2.111**
-
-
-1.816**
-
-
-
-
US
Core PCE QoQ - Preliminary
-0.360**
-0.689*
0.94
-
-1.427***
-
-0.408**
-
-
-
-1.096***
-
US
Durables Ex Transportation
-
-0.448***
-
-
-
-
-
-
-
-
-0.232*
-
US
Empire Manufacturing
-
0.473***
-
-
0.532***
-
-
0.418***
-
-
0.516***
-
US
Existing Home Sales
-
0.356**
-
-
0.414***
-
-0.178**
-0.495***
0.00
0.211**
-
-
US
GDP Annualized QoQ - Advance
-
-0.897***
-
-
-
-
-
-0.927***
-
-
-0.848***
-
US
GDP Annualized QoQ Preliminary
-
-0.897***
-
-
-1.011***
-
-
-0.649*
-
-
-1.201***
-
US
Housing Starts
-
0.735***
-
-
0.382***
-
-0.170*
0.586***
0.00
-
0.602***
-
0.29
Notes: Table 17 presents estimation results of the news surprise effect and unscheduled news effect on depth at 2nd to 5th levels and 5th to 10th levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
91
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB (continued)
Depth
Ask Side
2nd
CN
Scheduled News
US
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
Recession
Expansion
P-diff
Recession
5th to 10th level
Recession
Expansion
P-diff
Expansion
P-diff
Recession
Expansion
P-diff
IBD/TIPP Economic Optimism
-
-0.441**
-
-
-
-
-
-
-
-
-
-
US
Initial Jobless Claims
-
-0.205**
-
-
-0.403***
-
-
-0.308***
-
-
-0.446***
-
US
ISM Milwaukee
0.619***
-1.079***
0.00
0.413**
-
-
-
-
-
-
0.989***
-
US
New Home Sales
-0.477**
0.382***
0.00
-0.762***
0.523***
0.00
-0.506***
0.410***
0.02
-0.661***
0.485***
0.00
US
Nonfarm Productivity - Final
-
0.651**
-
-
-
-
-
-
-
-
-0.722**
-
US
Nonfarm Productivity - Preliminary
0.549***
-
-
0.524***
-
-
0.402**
-
-
0.387***
US
PCE Core MoM
-
-0.349***
-
-
-0.388***
-
-
-0.334**
-
-
-0.244**
-
US
Pending Home Sales MoM
-
-0.842***
-
-
-0.370**
-
0.285**
-0.856***
0.03
0.273**
-0.331**
0.00
US
Personal Spending
-
0.466***
-
0.357***
0.236*
0.01
-
0.332*
-
-
-
-
US
Philadelphia Fed Business Outlook
-
-0.318**
-
-
0.381**
-
-
-0.498***
-
-
-
-
US
PPI Ex Food and Energy MoM
0.214*
-0.394***
0.00
0.301**
-0.352***
0.02
-
0.521**
-
0.274**
-0.322***
0.00
US
PPI MoM
-0.189*
0.507***
0.04
-0.212*
0.567***
0.01
-
-0.272*
-
-
0.42**
-
US
Retail Sales Ex Auto MoM
-
-0.233*
-
-
-
-
-0.229**
-
-
-
-
-
US
Unemployment Rate
-0.250**
-
-
-
-
-
-
0.258*
-
-
-
-
US
Univ. of Michigan Confidence - Preliminary
-
0.244**
-
-
-
-
-
1.093*
-
-
-
-
US
Wholesale Inventories MoM
-
1.465***
-
-
0.675***
-
-
0.332***
-
-
0.859**
-
US
Personal Consumption - Preliminary
-
-
-
-
-
-
0.424**
-
-
-
-
-
-
Notes: Table 17 presents estimation results of the news surprise effect and unscheduled news effect on depth at 2nd to 5th levels and 5th to 10th levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
92
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB (continued)
Depth
Ask Side
2nd
to
5th
Bid Side
5th
level
to
10th
Recession
2nd to 5th level
level
Expansion
P-diff
Recession
5th to 10th level
CN
Scheduled News
Recession
Expansion
P-diff
Expansion
P-diff
Recession
Expansion
P-diff
US
Factory Orders
-
-
-
-
-0.264**
-
-
-
-
-
-
-
US
ISM Manufacturing
-
-
-
-
-
-
-
0.284*
-
-
0.242*
-
US
NAHB Housing Market Index
-
-
-
-
-0.266**
-
0.247*
-
-
-
-
-
US
Pending Home Sales MoM
-
-
-
-
-
-
-
-
-
-
-
-
US
Import Price Index MoM
-
-
-
-
0.866***
-
-
-
-
-
-
-
US
Industrial Production MoM
-
-
-
-
-0.644***
-
-
-
-
-
-
-
US
Trade Balance
-
-
-
-
0.46***
-
-
-
-
-
-
-
US
CPI Ex Food and Energy MoM
-
-
-
-
-
-
-
-
-
-
0.444**
-
US
ISM Non-Manf. Composite
-
-
-
-
-
-
-
-
-
0.351***
-0.449*
0.00
Panel C: European Countries
GE
Factory Orders WDA YoY - Preliminary
-
0.237**
-
-
-
-
-
-
-
-
-0.209**
-
GE
Industrial Production SA MoM - Preliminary
-
-0.424***
-
-0.129*
-
-
-
-0.6***
-
-
-0.27*
-
GE
Private Consumption QoQ
0.360*
-
-
1.029***
-0.666**
0.23
-
-
-
-
-0.455*
-
GE
ZEW Survey Current Situation
0.167*
-
-
-
0.506**
-
0.31***
-
-
0.204**
-
-
GE
ZEW Survey Expectations
-
-1.267***
-
-
-0.776***
-
-
-0.715***
-
-
-1.263***
-
GE
IFO Business Climate
-
-
-
-
0.248*
-
-
0.193**
0.301**
0.02
Notes: Table 17 presents estimation results of the news surprise effect and unscheduled news effect on depth at
2nd
to
0.212*
5th
levels and
5th
to
10th
levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
93
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 17. Robustness Results of Surprise on Depth at different levels in the LOB (continued)
Depth
Ask Side
2nd
CN
Scheduled News
GE
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Construction Investment QoQ
-
-
-
1.108**
-
-
-
-
-
-
-
-
GE
Exports QoQ
-
-
-
0.44**
1.348*
0.39
-
-
-
-
-
-
GE
Imports QoQ
-
-
-
-0.641***
-
-
-
-
-
-
-
-
GE
Unemployment Rate
-
-
-
-
-
-
-
-
-
0.139*
-
-
PO
CPI MoM
-
0.418*
-
-
-
-
-
-
-
-
-
-
PO
GDP YoY - Final
-
-
-
-
-
-
-
-
-
-0.369*
-
-
IT
GDP WDA QoQ - Preliminary
-
-0.878**
-
-
-
-
-
-
-
-
-
-
IT
Industrial Production WDA YoY
-
-
-
0.184*
-
-
-
-
-
0.192**
-0.296*
0.53
IT
Retail Sales MoM
-
-
-
-
-0.34**
-
-
-
-
-
-0.419**
-
IT
Trade Balance Total
-
-
-
-
-
-
-
-
-
-
0.337**
-
FR
Own-Company Production Outlook
-
0.459**
-
-
-
-
-
-
-
-
-
-
FR
Consumer Spending (MoM)
-
-
-
0.229*
-
-
-
-
-
-
-
-
SP
Retail Sales WDA YoY
-
-
-
-
-
-
0.298*
-
-
-
-0.622*
-
SP
CPI MoM
-
-
-
-
0.19*
-
-
-
-
-
-
-
SP
Unemployment Rate
-
-
-
-
0.764*
-
-
-
-
-0.551***
-
-
US
US Unscheduled News
-
-0.140***
-
-0.040***
-
-
-0.070***
-
-
-0.021***
-0.043***
0.00
EC
EC Unscheduled News
-0.033**
0.023***
0.00
-0.052***
0.076***
-
0.032**
0.00
-0.047***
-
-
Notes: Table 17 presents estimation results of the news surprise effect and unscheduled news effect on depth at
2nd
Recession
5th to 10th level
-0.066***
to
5th
levels and
Expansion
5th
to
10th
P-diff
Recession
Expansion
P-diff
levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
94
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 18. Robustness Results of Surprise on Slope at different levels in the LOB
Slope
Ask Side
2nd
CN
Scheduled News
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
5th to 10th level
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Panel A: Euro Zone News
EC
Gross Fix Cap QoQ - Preliminary
0.715*
-
-
-
-
-
1.215***
-
-
-
-
-
EC
Household Cons QoQ - Preliminary
-0.801*
0.498*
0.34
-
-
-
-1.049*
-
-
-0.746*
-
-
EC
GDP SA QoQ - Final
-
-
-
-
-
-
-0.628***
-
-
0.290*
-
-
EC
Govt Expend QoQ - Preliminary
-
-
-
-
-
-
0.933*
-
-
-
-
-
EC
PMI Manufacturing - Preliminary
-
-
-
-
-
-
0.248*
-
-
-0.176*
-
-
EC
Trade Balance SA
-
-
-
0.31**
-
-
-
0.514**
0.18**
-
-
EC
Labour Costs YoY
-
-
-
0.551*
-
-
-
-
-
-
-
-
EC
ZEW Survey Expectations
-
-
-
-0.344**
-
-
-
-
-
-
-0.316**
-
EC
CPI Estimate YoY
-
-
-
-
-
-
-
-
-
-
-0.521**
-
EC
Industrial Production SA MoM
-
-
-
-
-
-
-
-
-
-
-0.337*
-
0.159**
-0.962***
0.7
Panel B: US News
US
ADP Employment Change
-
-0.735*
-
-
-1.324***
US
Avg Weekly Hours Production
-0.259**
-
-
-0.318**
-
-
-0.259**
-
-
-0.316***
-
-
US
Chicago Purchasing Manager
0.231*
-
-
-
0.331**
-
-
-
-
-
0.339**
-
US
Construction Spending MoM
0.255**
-
-
0.349**
0.378*
0.02
-
-
-
0.265***
-
-
US
Consumer Confidence Index
-0.258**
-
-
-
-
-
-
-
-
-
Notes: Table 18 presents estimation results of the news surprise effect and unscheduled news effect on slope at
2nd
to
5th
levels and
5th
to
10th
levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
95
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 18. Robustness Results of Surprise on Slope at different levels in the LOB (continued)
Slope
Ask Side
2nd
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
5th to 10th level
CN
Scheduled News
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
US
Core PCE QoQ – Preliminary
0.371*
-0.997**
0.49
-
-0.963***
-
-
-1.749***
-
0.445***
-1.747***
0.02
US
Durables Ex Transportation
0.211**
-
-
-
-
-
-
-
-
-
-
-
US
Existing Home Sales
-
0.594***
-
-
-
-
-
-
-
-
0.414***
-
US
FOMC Rate Decision
-0.144*
-
-
-
-
-
-
-
-
-
-
-
US
GDP Annualized QoQ - Preliminary
0.644**
-
-
0.576**
-
-
-
-
-
0.955***
-0.971***
0.06
US
Initial Jobless Claims
-
-0.213*
-
-
-0.287***
-
-
-0.319***
-
-
-0.275***
-
US
ISM Milwaukee
-
0.825*
-
-0.341*
1.171***
0.05
-
-
-
-
1.253***
-
US
ISM Non-Manf. Composite
-
-0.979*
-
-
-1.481***
-
-
-2.273***
-
-
-0.956***
-
US
New Home Sales
-
0.398**
-
-
-0.267**
-
-
-
-
-
-
-
US
Nonfarm Productivity - Preliminary
0.535***
-
-
-
0.443***
-
0.333*
-
-
0.503***
0.338***
0.18
US
PCE Core MoM
-
-0.550***
-
-
-0.218*
-
-
-
-
0.208*
-0.392***
0.1
US
Personal Consumption - Preliminary
-0.528*
-
-
-0.564**
-
-
-
-
-
-0.903***
-
-
US
Personal Spending
-
0.39*
-
-
0.246*
-
-
0.479**
-
-
0.346**
-
US
PPI Ex Food and Energy MoM
0.424*
-0.628***
0.00
0.33***
-0.312*
0.01
-
-
-
-
-
-
US
PPI MoM
-
0.82***
-
-
-
-
0.263*
-
-
0.372***
0.52***
0.15
US
Trade Balance
0.205**
-
-
0.258**
-
-
-
-
-
0.266***
-
-
US
Unemployment Rate
-
0.61**
-
-
-
-
-
0.759*
-
-
-
-
Notes: Table 18 presents estimation results of the news surprise effect and unscheduled news effect on slope at 2nd to 5th levels and 5th to 10th levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
96
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 18. Robustness Results of Surprise on Slope at different levels in the LOB (continued)
Slope
Ask Side
2nd
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
5th to 10th level
CN
Scheduled News
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
US
Import Price Index MoM
-
-
-
-
0.274*
-
0.217**
-
-
0.261***
-
-
US
Net Long-term TIC Flows
-
-
-
-0.204**
0.27*
0.07
-0.314**
0.485**
0.02
-0.176*
0.271*
0.03
US
Business Inventories
-
-
-
-0.167*
0.327**
0.28
-
-
-
-0.165*
-
-
US
Change in Nonfarm Payrolls
-
-
-
0.29***
-
-
0.239*
-0.467**
0.05
0.173*
-
-
US
Housing Starts
-
-
-
-
0.267**
-
-
-
-
-
-
-
US
IBD/TIPP Economic Optimism
-
-
-
-
0.419**
-
-
-
-
-
-
-
US
ISM Manufacturing
-
-
-
0.232**
0.292**
0.00
-
-
-
-
-
-
US
Nonfarm Productivity – Final
-
-
-
-
-0.718**
-
-
-
-
0.218**
-
-
US
Pending Home Sales MoM
-
-
-
0.256**
-
-
-
-0.494*
-
0.324***
0.358**
0.03
US
Philadelphia Fed Business Outlook
-
-
-
-0.166**
-
-
-
-
-
-0.154*
0.264*
0.27
US
Retail Sales Ex Auto MoM
-
-
-
-
-
-
-
-
-
0.175**
-
-
US
Univ. of Michigan Confidence - Preliminary
-
-
-
-
-
-
-0.262**
-
-
-
0.368***
-
Panel C: European Countries News
GE
Imports QoQ
-
0.735*
-
-
-
-
-
0.735*
-
-
-
-
GE
Industrial Production SA MoM - Preliminary
-
0.529**
-
-
0.274*
-
-
0.529**
-
-
-
-
GE
Retail Sales MoM
-
-1.57***
-
-
-
-
-0.224***
-
-
-0.17*
-0.235**
0.25
GE
GDP SA QoQ - Preliminary
-
-
-
0.196*
-
-
-
-
-
-
-
-
Notes: Table 18 presents estimation results of the news surprise effect and unscheduled news effect on slope at
2nd
to
5th
levels and
5th
to
10th
levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
97
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Table 18. Robustness Results of Surprise on Slope at different levels in the LOB (continued)
Slope
Ask Side
2nd
to
5th
Bid Side
5th
level
to
10th
2nd to 5th level
level
5th to 10th level
CN
Scheduled News
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
Recession
Expansion
P-diff
GE
PPI MoM
-
-
-
-0.125*
-
-
-
-
-
-
-
-
SP
Retail Sales WDA YoY
-
1.57***
-
-
-
-
-
-
-
-
-
-
SP
Unemployment Rate
-0.484**
-
-
-
-
-
-
-
-
-
-
-
SP
CPI MoM
-
-
-
0.172*
-
-
-
-
-
-
-
-
IT
Industrial Production WDA YoY
-
-0.408*
-
-
-
-
-
-
-
-
-
-
IT
Retail Sales MoM
-
-
-
0.308***
-
-
0.365***
-
-
0.195*
-0.487***
0.01
FR
PPI MoM
-0.172**
-
-
-
-
-
-
-
-
-
-
-
FR
Consumer Spending (MoM)
-
-
-
-
-
-
-
-
-
-0.215*
-
-
Panel D: Unscheduled News
US
US Unscheduled News
0.014***
0.146***
0.00
-0.007**
0.091***
0.01
-
0.103***
-
0.017***
0.039***
0.00
EC
EC Unscheduled News
-0.056***
0.051***
0.00
-0.02***
-
-
-0.030***
0.098***
0.02
-0.023***
0.039***
0.00
Notes: Table 18 presents estimation results of the news surprise effect and unscheduled news effect on slope at 2nd to 5th levels and 5th to 10th levels in the LOB at ask side and bid side. CN
presents the corresponding country name of the news: EC - Euro Zone; GE - German; US – United States; FR - France; IT - Italy; PO - Poland; SP - Spain. *, **, *** denotes the prob. of
insignificance of news are at 1%, 5% and 10% levels respectively. P_diff presents the difference between the coefficients in expansion and recession. A significant value of P_diff indicate that the
coefficients are statistically different over expansion and recession.
98
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figures
Figure 1. Intraday Pattern of Characteristics
a. Intraday Pattern of Depth
350,000,000
b. Intraday Pattern of Quoted Spread
2.5
300,000,000
2
250,000,000
200,000,000
1.5
150,000,000
1
100,000,000
0.5
50,000,000
0
0:05
1:15
2:25
3:35
4:45
5:55
7:05
8:15
9:25
10:35
11:45
12:55
14:05
15:15
16:25
18:40
19:50
21:00
22:10
23:20
0:05
1:25
2:45
4:05
5:25
6:45
8:05
9:25
10:45
12:05
13:25
14:45
16:05
18:30
19:50
21:10
22:30
23:50
0
c. Intraday Pattern of Slope
d. Intraday Pattern of Volatility
0.07
0.06
40,000
35,000
0.05
30,000
0.04
25,000
0.03
20,000
15,000
0.02
10,000
0.01
5,000
0:05
1:10
2:15
3:20
4:25
5:30
6:35
7:40
8:45
9:50
10:55
12:00
13:05
14:10
15:15
16:20
18:30
19:35
20:40
21:45
22:50
23:55
0:05
1:15
2:25
3:35
4:45
5:55
7:05
8:15
9:25
10:35
11:45
12:55
14:05
15:15
16:25
18:40
19:50
21:00
22:10
23:20
0
0
Notes: Figure 1 presents the intraday patterns of depth, quote spread, slope and volatility from Jan 3, 2006 to Dec 31,
2009. For each graph, the x-axis is the 275 intervals in a trading day, and the title displays the name of the variable
depicted. Un-weighted averages across all intervals in one day are shown. All variables are drawn without adjustment
for intraday seasonality.
99
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figure 2. Intraday Announcement Cluster
Euro Zone Countries
700
600
500
400
300
200
100
9:45
9:45
14:45
9:00
9:00
14:10
8:30
8:30
12:00
7:45
7:45
11:00
6:30
6:30
5:30
5:00
4:30
3:55
3:45
3:25
2:55
2:45
2:00
1:00
0
US
1200
1000
800
600
400
200
14:45
14:10
12:00
11:00
5:30
5:00
4:30
3:55
3:45
3:25
2:55
2:45
2:00
1:00
0
Notes: Figure 2 plots the bar charts for the cumulated macroeconomic news announcements
frequencies from Jan 3, 2006 to Dec 31, 2009. The news included here are the total number of valid
news filtered by the first round of filtered introduced in section 4.1.3. The Vertical Axis is the
number of announcements. The Horizontal Axis is the time a news announced stamped to minutes.
100
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figure 3. Transition Variable ISM
60
55
50
45
40
35
9/1/2009
11/1/2009
7/1/2009
5/1/2009
1/1/2009
3/1/2009
11/1/2008
9/1/2008
7/1/2008
5/1/2008
3/1/2008
1/1/2008
11/1/2007
9/1/2007
7/1/2007
5/1/2007
1/1/2007
3/1/2007
9/1/2006
11/1/2006
7/1/2006
5/1/2006
1/1/2006
3/1/2006
30
Notes: Figure 3 plots regime indicator, ISM from 2006 to 2009. ISM (Institute of Supply
Management) is manufacturing index for US business cycles. The value of 50 means that half of
the survey participants believe the economy is in good state and half think it is bad state. ISM
below 50 indicates worse economy condition. The Vertical Axis is the magnitude of ISM.
101
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figure 4. Estimation Results of Logistic Transition Function
G
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
2008/11/1
2009/1/1
2009/3/1
2009/5/1
2009/7/1
2008/11/1
2009/1/1
2009/3/1
2009/5/1
2009/7/1
2009/9/1
2008/9/1
2008/9/1
2009/11/1
2008/7/1
2008/7/1
2009/9/1
2008/5/1
2008/5/1
2009/11/1
2008/3/1
2008/3/1
2008/1/1
2007/9/1
2007/11/1
2007/7/1
2007/5/1
2007/1/1
2007/3/1
2006/9/1
2006/11/1
2006/7/1
2006/5/1
2006/1/1
2006/3/1
0
NBER
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
2008/1/1
2007/9/1
2007/11/1
2007/7/1
2007/5/1
2007/1/1
2007/3/1
2006/11/1
2006/9/1
2006/7/1
2006/5/1
2006/1/1
2006/3/1
0
Notes: Figure 4 plots the fitted G from (1) and NBER dates from 2006 to 2009. G is between
0 (lower regime: recession) and 1 (higher regime: expansion). NBER is 1 when economy is
in expansion; NBER is 0 which indicates economy is in recession.
102
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figure 5. Intraday Patterns of Alternative Characteristics
a. Intraday Pattern of NORMSLOPE
b. Intraday Pattern of WSLOPE
1.6
120,000
1.4
100,000
1.2
80,000
1
60,000
0.8
0.6
40,000
0.4
0
0:05
1:10
2:15
3:20
4:25
5:30
6:35
7:40
8:45
9:50
10:55
12:00
13:05
14:10
15:15
16:20
18:30
19:35
20:40
21:45
22:50
23:55
0.2
0
0:05
1:15
2:25
3:35
4:45
5:55
7:05
8:15
9:25
10:35
11:45
12:55
14:05
15:15
16:25
18:40
19:50
21:00
22:10
23:20
20,000
c. Intraday Pattern of SIZESPREAD
d. Intraday Pattern of Volatility
0.07
0.25
0.06
0.2
0.05
0.15
0.04
0.03
0.1
0.02
0.05
0.01
0:05
1:10
2:15
3:20
4:25
5:30
6:35
7:40
8:45
9:50
10:55
12:00
13:05
14:10
15:15
16:20
18:30
19:35
20:40
21:45
22:50
23:55
0
0:05
1:10
2:15
3:20
4:25
5:30
6:35
7:40
8:45
9:50
10:55
12:00
13:05
14:10
15:15
16:20
18:30
19:35
20:40
21:45
22:50
23:55
0
’
e. Intraday Pattern of SIZE
350,000,000
300,000,000
250,000,000
200,000,000
150,000,000
100,000,000
50,000,000
0:05
1:20
2:35
3:50
5:05
6:20
7:35
8:50
10:05
11:20
12:35
13:50
15:05
16:20
18:40
19:55
21:10
22:25
23:40
0
Notes: Figure 5 presents the intraday patterns of alternative measures of slope (NORMSLOPE and WSLOPE), depth
(SIZE) and spread (SIZESPREAD) from 2006 to 2009. In each graph, the x-axis the 5-min interval trading periods of a
trading day, while the title displays the name of the variable depicted. Un-weighted averages across all intervals in one
day are shown. All variables are not adjusted for intraday seasonality.
103
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Figure 6. Autocorrelation Coefficients of Log Transformed Filtered Volatility
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1
251
501
751
1001
1251
-0.05
0.3
0.25
0.2
0.15
0.1
0.05
0
1
251
501
751
1001
1251
-0.05
Notes: Figure 6 shows the correlogram of autocorrelation coefficients of two
method of volatility with total lag 1400 intervals which contains five days of
since each day contain 275 intervals. The dashed line in the above figure
shows the correlogram of 5-min absolute returns, Abs_return and its
corresponding log-transformed filtered volatility, Volat. The dashed line in
the below figure shows the correlogram of 5-min absolute returns, Abs_ret,
and its corresponding log-transformed filtered volatility, Y. Vertical axis
shows the magnitude of autocorrelation coefficients.
104
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Appendix A
In the Appendix A, we give an example of VAR-STR model for alternative measures of
characteristics. For the category “slope”, we have three measures of “slope”, which are
introduced in section 3; “NORM SLOPE” is calculated with the same logic of “slope” except
that the “NORM SLOPE” for a tick is normalized with regard of the total size on that tick.
The third slope measures is size-weighted slope by Kozhan and Salmon (2010). For the
category “spread”, we have two measures of “spread”: “quoted spread” introduced in section
3 and size-weighted spread. Also, we have two measures of “volatility” with respect to two
measures of return. One is return is calculated by size-weighted price introduced in section 3,
the other is the return calculate by price.
For example, in the case of “NORM SLOPE”, volatility based on the best quote (6.1.4),
the size (6.1.1) and the size-weighted spread, the VAR-STR model with exogenous variables
news surprise 𝑆𝑞 is:
𝐽
𝑄
𝑈𝑆
𝐸𝐶
Ω𝑡,𝑛 = 𝛼𝑡,𝑛 + ∑𝑗 𝛽𝑗 Ω𝑡,𝑛−𝑗 + 𝜆𝐴𝑉𝑡,𝑛 + ∑𝑞=1 𝜃𝑞 S𝑞,𝑡,𝑛 + 𝜂1 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+ 𝜂2 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+
𝑈𝑆
𝐸𝐶 ̂
′
{𝛼𝑡,𝑛
+ ∑𝑄𝑞=1 𝜃𝑞′ S𝑞,𝑡,𝑛 + 𝜂1′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
+ 𝜂2′ 𝑢𝑛𝑠𝑐ℎ𝑡,𝑛
}𝐺 (𝜓𝑡,𝑛 , 𝛾, 𝑐) + 𝜀𝑡,𝑛
(14)
Similar as the case of equation (8), the vector of endogenous variable in (13) is: Ω𝑡,𝑛 =
′
𝑁𝑂𝑅𝑀
(𝐷𝑒𝑝𝑡ℎ𝑡,𝑛 , 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛
, 𝑊𝑆𝑃𝑅𝐸𝐴𝐷𝑡,𝑛 , 𝑉𝑜𝑙𝑎𝑡𝑡,𝑛 ) , where 𝑊𝑆𝑃𝑅𝐸𝐴𝐷𝑡,𝑛 is the size-weighted
𝑁𝑂𝑅𝑀
spread at interval n on day t; 𝑆𝑙𝑜𝑝𝑒𝑡,𝑛
is the NORM SLOPE at interval n on day t; And
𝐴𝑉𝑡,𝑛 is the seasonality dummy of quoted depth, WSPREAD and NORM SLOPE
𝑑𝑒𝑝𝑡ℎ
𝑤𝑠𝑝𝑟𝑒𝑎𝑑
𝑛𝑜𝑟𝑚𝑠𝑙𝑜𝑝𝑒
are 𝐴𝑉𝑡,𝑛
, 𝐴𝑉𝑡,𝑛
or 𝐴𝑉𝑡,𝑛
respectively.
105
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Appendix B
Appendix B shows the additional data for equation (7) and (11).
1. Daily Exchange Rate
In the section 3.4.2, the daily spot exchange rate return data was needed beyond our
sample range to construct one day ahead volatility component in FFF equation. The sample
of daily spot exchange rates was from the initial year of the euro by using Bloomberg HP
(Historical Price) function.
2. Consolidated Macroeconomic news variable
Consolidated Macroeconomic News vector is used to obtain the fitted transaction
variable in equation (2).
Before the polynomial structure, regardless of country and
category, we construct a dummy which equals one as long as news occurs, otherwise the
dummy is zero. Then we construct a third order polynomial structure to create a vector that
can capture the decay impact on volatility within two hours (Andersen et al., 2003):
𝑛 3
𝑛 2
𝑛
𝜌(𝑛) = 𝑐0 (1 − ( 𝐼 ) )+𝑐1 (1 − ( 𝐼 ) ) 𝑛 + 𝑐2 (1 − 𝐼 ) 𝑛2 ,
(15)
where response window n=1….25 is the number of interval. And n=25 is the sum intervals of
two hours (5-min interval). 𝜌(𝑛) describes the decay pattern for the effect of news on
volatility. 𝜌(𝑛) is the fitted values corresponding to the difference between the average
absolute return at each time interval just after the news announcements and the average
absolute return computed for the whole sample .
106
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Appendix C
Name list of news and country is shown in the Appendix C. Advance/Preliminary/Final denotes Advance,
Preliminary and Final report for a certain news that announced for several time with revision of the figures.
Effect describe the standard to classify the news in to good or bad. SA/NSA means Seasonal Adjusted or Non
Seasonal Adjusted figures. YoY, MoM, QoQ denotes the comparison between the current released figure and
the previous figure Year over Year, Month over Month, Quarter over Quarter. WDA denotes for Weighted
Density Approximation.15
15
Country
News
Effect
EC
Business Climate Indicator
Actual > Forecast = Good News
EC
CPI Core YoY -Final
Actual > Forecast = Good News
EC
CPI Estimate YoY
Actual > Forecast = Good News
EC
ECB Announces Interest Rates
Actual > Forecast = Good News
EC
GDP SA QoQ -Final
Actual > Forecast = Good News
EC
Govt Expend QoQ -Preliminary
Actual > Forecast = Good News
EC
Gross Fix Cap QoQ -Final
Actual > Forecast = Good News
EC
Gross Fix Cap QoQ -Preliminary
Actual > Forecast = Good News
EC
Household Cons QoQ -Preliminary
Actual > Forecast = Good News
EC
Industrial New Orders SA (MoM)
Actual > Forecast = Good News
EC
Industrial Production SA MoM
Actual > Forecast = Good News
EC
Labour Costs YoY
Actual > Forecast = Good News
EC
PMI Manufacturing -Preliminary
Actual > Forecast = Good News
EC
Retail Sales MoM
Actual > Forecast = Good News
EC
Trade Balance SA
Actual > Forecast = Good News
EC
ZEW Survey Expectations
Actual > Forecast = Good News
US
ADP Employment Change
Actual > Forecast = Good News
US
Avg Hourly Earning MOM Prod
Actual > Forecast = Good News
US
Avg Weekly Hours Production
Actual > Forecast = Good News
The source are Bloomberg and www.Forexfactory.com
107
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
US
Business Inventories
Actual < Forecast = Good News
US
Change in Nonfarm Payrolls
Actual > Forecast = Good News
US
Chicago Purchasing Manager
Actual > Forecast = Good News
US
Construction Spending MoM
Actual > Forecast = Good News
US
Consumer Confidence Index
Actual > Forecast = Good News
US
Core PCE QoQ -Advance
Actual > Forecast = Good News
US
Core PCE QoQ -Preliminary
Actual > Forecast = Good News
US
CPI Ex Food and Energy MoM
Actual > Forecast = Good News
US
Durables Ex Transportation
Actual > Forecast = Good News
US
FOMC Rate Decision
Actual > Forecast = Good News
US
Empire Manufacturing
Actual > Forecast = Good News
US
Factory Orders
Actual > Forecast = Good News
US
Existing Home Sales
Actual > Forecast = Good News
US
GDP Annualized QoQ -Advance
Actual > Forecast = Good News
US
GDP Annualized QoQ -Preliminary
Actual > Forecast = Good News
US
Housing Starts
Actual > Forecast = Good News
US
IBD/TIPP Economic Optimism
Actual > Forecast = Good News
US
Import Price Index MoM
Actual > Forecast = Good News
US
Industrial Production MoM
Actual > Forecast = Good News
US
Initial Jobless Claims
Actual < Forecast = Good News
US
ISM Manufacturing
Actual > Forecast = Good News
US
ISM Milwaukee
Actual > Forecast = Good News
US
ISM Non-Manf. Composite
Actual > Forecast = Good News
US
Net Long-term TIC Flows
Actual > Forecast = Good News
US
NAHB Housing Market Index
Actual > Forecast = Good News
US
New Home Sales
Actual > Forecast = Good News
US
Nonfarm Productivity -Final
Actual > Forecast = Good News
US
Nonfarm Productivity -Preliminary
Actual > Forecast = Good News
108
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
US
PCE Core MoM
Actual > Forecast = Good News
US
Pending Home Sales MoM
Actual > Forecast = Good News
US
Personal Consumption -Preliminary
Actual > Forecast = Good News
US
Personal Spending
Actual > Forecast = Good News
US
Philadelphia Fed Business Outlook
Actual > Forecast = Good News
US
PPI Ex Food and Energy MoM
Actual > Forecast = Good News
US
PPI MoM
Actual > Forecast = Good News
US
Retail Sales Ex Auto MoM
Actual > Forecast = Good News
US
Trade Balance
Actual > Forecast = Good News
US
Unemployment Rate
Actual < Forecast = Good News
US
Univ. of Michigan Confidence -Preliminary
Actual > Forecast = Good News
US
Wholesale Inventories MoM
Actual < Forecast = Good News
SP
CPI EU Harmonised YoY -Final
Actual > Forecast = Good News
SP
CPI MoM
Actual > Forecast = Good News
SP
Retail Sales WDA YoY
Actual > Forecast = Good News
SP
Unemployment Rate
Actual < Forecast = Good News
PO
CPI MoM
Actual > Forecast = Good News
PO
GDP YoY -Final
Actual > Forecast = Good News
IT
Business Confidence
Actual > Forecast = Good News
IT
GDP WDA QoQ -Final
Actual > Forecast = Good News
IT
Industrial Production WDA YoY
Actual > Forecast = Good News
IT
Retail Sales MoM
Actual > Forecast = Good News
IT
Total investments
Actual > Forecast = Good News
IT
Trade Balance Total
Actual > Forecast = Good News
IT
Unemployment Rate Quarterly
Actual < Forecast = Good News
GE
Construction Investment QoQ
Actual > Forecast = Good News
GE
Exports QoQ
Actual > Forecast = Good News
109
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
GE
Factory Orders WDA YoY -Preliminary
Actual > Forecast = Good News
GE
GDP SA QoQ -Preliminary
Actual > Forecast = Good News
GE
IFO Business Climate
Actual > Forecast = Good News
GE
Imports QoQ
Actual > Forecast = Good News
GE
Industrial Production SA MoM -Preliminary
Actual > Forecast = Good News
GE
PPI MoM
Actual > Forecast = Good News
GE
Private Consumption QoQ
Actual > Forecast = Good News
GE
Retail Sales MoM
Actual > Forecast = Good News
GE
Unemployment Rate
Actual < Forecast = Good News
GE
ZEW Survey Current Situation
Actual > Forecast = Good News
GE
ZEW Survey Expectations
Actual > Forecast = Good News
FR
Consumer Spending (MoM)
Actual > Forecast = Good News
FR
Own-Company Production Outlook
Actual > Forecast = Good News
FR
PPI MoM
Actual > Forecast = Good News
110
GOODMAN SCHOOL OF BUSINESS, BROCK UNIVERSITY
YUSI TAO
Appendix D
Appendix D shows the diagram of LOB.
Diagram 1
l=1…L
Ask Price
Ask size
l=4
Interval 2
Ask Side
l=3
l=2
l=1
Tick 3 n=2
Tick 2
n=1
Tick 1
𝜏3
l=1
l=2
l=3
l=1…L
Time/ day t
𝜏2
Bid Price
Bid Size
Bid Side
𝜏1
111